<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by BioTuring Team on Medium]]></title>
        <description><![CDATA[Stories by BioTuring Team on Medium]]></description>
        <link>https://medium.com/@bioturing?source=rss-64e2fc4336f3------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*2IcxeIe3ZCCyArldI4FFXg.png</url>
            <title>Stories by BioTuring Team on Medium</title>
            <link>https://medium.com/@bioturing?source=rss-64e2fc4336f3------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 03:43:45 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@bioturing/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Getting back to our dream. Pushing the boundary of knowledge.]]></title>
            <link>https://bioturing.medium.com/getting-back-to-our-dream-pushing-the-boundary-of-knowledge-e0271d121402?source=rss-64e2fc4336f3------2</link>
            <guid isPermaLink="false">https://medium.com/p/e0271d121402</guid>
            <category><![CDATA[coming-back]]></category>
            <dc:creator><![CDATA[BioTuring Team]]></dc:creator>
            <pubDate>Sun, 06 Jul 2025 03:19:54 GMT</pubDate>
            <atom:updated>2025-07-06T03:19:54.140Z</atom:updated>
            <content:encoded><![CDATA[<p>Getting back to our dream. Pushing the boundary of knowledge.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e0271d121402" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Why, When and How of 3D PCA]]></title>
            <link>https://bioturing.medium.com/the-why-when-and-how-of-3d-pca-bdb5c209f693?source=rss-64e2fc4336f3------2</link>
            <guid isPermaLink="false">https://medium.com/p/bdb5c209f693</guid>
            <category><![CDATA[zoology]]></category>
            <category><![CDATA[data-visualization]]></category>
            <category><![CDATA[scatter-plots]]></category>
            <category><![CDATA[plants]]></category>
            <category><![CDATA[biology]]></category>
            <dc:creator><![CDATA[BioTuring Team]]></dc:creator>
            <pubDate>Thu, 15 Oct 2020 13:34:18 GMT</pubDate>
            <atom:updated>2020-10-18T07:44:25.859Z</atom:updated>
            <content:encoded><![CDATA[<p>Principal component analysis (PCA) is one of the most popular dimension reduction methods. It works by converting the information in a complex dataset into <em>principal components </em>(PC), a few of which can describe most of the variation in the original dataset. The data can then be plotted with just the two or three most descriptive PCs, producing a 2D or 3D scatter plot. This PCA plot makes it possible to visualize strong patterns, such as groups of similar observations, in the original dataset.</p><p>Read more: <a href="https://blog.bioturing.com/2018/06/14/principal-component-analysis-explained-simply/">Principal component analysis explained simply</a>.</p><p><strong>In this post, we will focus on 3D PCA: what it is, when to use it, and how to run 3D PCA using BioVinci.</strong></p><h3>What is 3D PCA?</h3><p>Most of the time, a PCA plot is a 2D scatter plot in which the data is plotted with two most descriptive principal components. However, you can choose to plot with three PCs instead, and this will create a 3D scatter plot, also called 3D PCA.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/686/0*HS0-LPBVEHoVQW9Q.png" /></figure><p><em>3D PCA created with BioVinci</em></p><h3>When should I use 3D PCA?</h3><p>The key difference between 2D PCA and 3D PCA is the <em>number of principal components</em> being selected for plotting. In PCA, principal components are constructed to capture the most variation in the dataset: PC1 describes the most variation, PC2 describes the second most variation, and so forth. As a result, the first two or three PCs can capture most of the variation and the rest can be discarded without losing much information. This can be seen in a scree plot.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/897/0*_t5hHNaemru5_bev.png" /></figure><p><em>A scree plot shows how much variation each PC captures from the data. The y axis represents eigenvalues, i.e. the amount of variation. The x axis represents numbers of principal components.</em></p><p>Read more: <a href="https://blog.bioturing.com/2018/06/18/how-to-read-pca-biplots-and-scree-plots/">How to read PCA biplots and scree plot</a></p><h3>1. Cattell’s scree test: select PCs before the “elbow” in the scree plot</h3><p>A scree plot provides a good indication whether or not you should select three principal components to plot, thus creating a 3D PCA. A good scree curve usually has a bend (“elbow”) that can be used as the cutoff point for PC selection. The PCs before the “elbow” are significant and should be kept; while the PCs after the bend could be discarded without losing much information.</p><p>If the scree plot bends after the first two PCs, those two should be kept for plotting and a 2D PCA is sufficient to describe the data. If the bend occurs after three PCs instead, that is a call for 3D PCA.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/1*O8kv6jVpmH9xr5Gk5miMkg.png" /><figcaption><em>Cattell’s scree test helps determine the number of PCs to be selected. Scree curve 1 bends at 2 PCs. Scree plot 2 has an “elbow” at 3 PCs, indicating that 3 PCs should be selected to plot a 3D PCA. Scree plots created with BioVinci.</em></figcaption></figure><h3>2. Kaiser’s rule: select PCs that have eigenvalues of at least 1</h3><p>Another rule for picking PCs is Kaiser’s rule, which states that the selected PCs should have eigenvalues of at least 1. A lower eigenvalue indicates that the PC contains less information than a single variable and therefore could be discarded.</p><p>Read more: <a href="https://www.researchgate.net/publication/280794624_Comparison_of_Five_Rules_of_Determining_the_Number_of_Components_to_Retain">Comparison of Five Rules of Determining the Number of Components to Retain</a></p><h3>3. Proportion of variance: select PCs that describe most of the variance</h3><p>Keep in mind that PCA helps reduce the overwhelming number of dimensions while still capturing the essence of the data. The proportion of variance plot shows how much variation is captured by each PC. What percentage of variance is considered “essential”? That is entirely up to you. The general recommendation is to select the PCs that can describe at least 70 to 80 percent of the variance. For example, if the first two PCs add up to 60% of variance, but including PC3 brings it up to 75%, then you might want to consider 3D PCA to avoid losing information.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*mTGu2SWAhdAVv6Tz.png" /><figcaption><em>Scree plots created with BioVinci.</em></figcaption></figure><p>4. Visualization: select PCs that best visualize patterns of the original dataset</p><p>Though PCA is not a clustering method, by reducing dimensionality, it can help visualize patterns, such as groups of similar expression profiles. These patterns might not be visible on a 2D PCA plot, but show up more clearly in 3D. In that case, 3D PCA could be what you need. Below are 2D and 3D PCA created by BioVinci, in which 3D PCA show clearer clustering.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/686/0*ZQ5nqA--MpToKXAZ.png" /><figcaption>3D PCA created by BioVinci</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/680/0*345V5gtuNuqKvbeQ.png" /><figcaption>2D PCA Scatter Plot created by BioVinci</figcaption></figure><h3><strong>How to run 3D PCA with BioVinci</strong></h3><p>If you know some coding, there are packages to create 3D PCA plot in R, Python. 3D Scatter Plot in Matplotlib can also plot 3D PCA.</p><p>If you are looking for a quick and easy option to run 3D PCA, try BioVinci. Our software packs powerful tools for data visualization and analysis with a very user-friendly interface.</p><h3>Follow these 4 easy steps to run 3D PCA with BioVinci:</h3><ul><li>Step 1. Import your data: Click <strong><em>Add New Workset</em></strong> and upload your data in our supported format ( .tsv, .csv, .xlsx, .feather)</li><li>Step 2. Select the <strong><em>Dimensionality reduction</em></strong> tab. Under <strong><em>Method</em>,</strong> select <strong><em>Principal component analysis</em></strong>.</li><li>Step 3. BioVinci will automatically run a 2D PCA. To change to 3D PCA, click the <strong><em>Option</em></strong> button on the top right corner, then select <strong><em>Basic chart</em></strong>. Drag the PC3 column and drop it into the <strong><em>Z axis box</em></strong>. Drag your grouping data to the <strong><em>Shape by box</em></strong> to have different shapes for your grouping.</li><li>Step 4. Click the <strong><em>Download</em></strong> button to save your plot as .sgv or .png.</li></ul><h4>Step 1: Import your data: Click Add New Workset and upload your data in our supported format ( .tsv, .csv, .xlsx, .feather)</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*hIXLfz8l5ne42bLn.png" /></figure><h4>Step 2: Select the Dimensionality reduction tab. Under Method, select Principal component analysis.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*qwHURsceFOQKE5_v.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*xJhyikPKjnFlZNqb.png" /></figure><h4>Step 3: BioVinci will automatically run a 2D PCA. To change to 3D PCA, click the Option button on the top right corner, then select Basic chart.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*GQKIVL3YT6coRyRK.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*XT-LTB1en3F6gK0o.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*exzRbcw0t9K05_BD.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*JFsg71wpHAojf0xJ.png" /></figure><h4>Step 4: Click Edit plot to change the legend and Click the Download button to save your plot as .sgv or .png.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*3yR7vUVLs4UQbWZ-.png" /></figure><p>Make your own 3D PCA plot with <a href="http://vinci.bioturing.com/">BioVinci</a> and share with us the result. There is always room to improve so please also share with us how to make this feature even better for you!</p><p>See our original post and check out other blog posts <a href="https://blog.bioturing.com/2020/10/12/the-why-when-and-how-of-3d-pca/">here</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bdb5c209f693" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[6 best box and whisker plot makers]]></title>
            <link>https://bioturing.medium.com/6-best-box-and-whisker-plot-makers-efdb2acc3952?source=rss-64e2fc4336f3------2</link>
            <guid isPermaLink="false">https://medium.com/p/efdb2acc3952</guid>
            <category><![CDATA[modified-box-plot]]></category>
            <category><![CDATA[r]]></category>
            <category><![CDATA[box-plot]]></category>
            <category><![CDATA[data-visualization-tools]]></category>
            <category><![CDATA[data-visualization]]></category>
            <dc:creator><![CDATA[BioTuring Team]]></dc:creator>
            <pubDate>Sun, 27 Sep 2020 14:03:05 GMT</pubDate>
            <atom:updated>2020-09-27T14:38:45.124Z</atom:updated>
            <content:encoded><![CDATA[<p>A <strong>box and whisker plot</strong>, or just <strong>box plot</strong>, is a graph that visualizes how spread-out a dataset is. A box plot displays 5 values: minimum, first quartile, median, third quartile, and maximum.</p><p><strong><em>How to read a box plot:</em></strong></p><p>The box is drawn from the first quartile to the third quartile. The line inside the box represents the median. The whiskers extend from both ends of the box to the minimum and maximum values. So a box and whisker plot is a type of distribution graph — it shows how the data is dispersed around the median, if the data is skewed, and whether or not it is symmetrical. Box plots take up very little space and therefore are a great option when you want to compare distributions between many groups.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/680/0*EaQinH5CEH8FOdbX.png" /></figure><p><strong><em>$Here we round up the six best box and whisker plot makers for you to choose from</em></strong>. They range from Python libraries for advanced users to easy drag-and-drop visualization apps that require no learning curve at all. These box and whisker plot makers can handle large datasets, and most of them are free or come with a free trial period.</p><p>A summary of the 6 best box and whisker plot makers:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*x4Mrk6Tm0TCMTSNP.png" /></figure><h3>User-friendly box and whisker plot makers</h3><h3>1.BioVinci</h3><p><a href="https://vinci.bioturing.com/">BioVinci</a> is a terrific software for data visualization and analysis. Making box plots with BioVinci is a cinch: import your data, drag and drop columns you want to plot, and you’re done. BioVinci is perhaps the only tool that does not require you to format your data in any particular way. The software also comes with a full suite of options to customize your box and whisker plots. Here’s a step-by-step guide on <a href="https://blog.bioturing.com/2018/05/13/how-to-draw-a-box-and-whisker-plot-with-outliers/">how to make box plots</a> with BioVinci, though you probably won’t need it — BioVinci is really intuitive.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*AUY2EI2weTJFNPew.png" /></figure><p>BioVinci is a rare option that offers a friendly interface without sacrificing power. Other than boxplots, the software is also capable of creating highly attractive graphs and running various analyses for data exploration such as PCA, feature selection, and heatmaps. While it requires a paid license, BioVinci offers a 15-day free trial for new users. Try BioVinci’s <a href="https://vinci.bioturing.com/">box and whisker plot maker</a> here.</p><p><strong>Pros:</strong> produces highly attractive plots, customizable, very user friendly.</p><p><strong>Cons:</strong> paid option (15 day free trial available).</p><h3>2. PAST</h3><p><a href="http://priede.bf.lu.lv/ftp/pub/TIS/datu_analiize/PAST/2.17c/download.html">PAST</a> is an open source statistical analysis software for Windows. PAST allows you to create a dozen types of graphs, including box plots, and it also lets you run various statistical tests.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/307/0*q0WFhgGgUOyIuD5g.png" /></figure><p>You can import data into PAST in XLS, TXT, or DAT formats. Creating box plots in PAST is pretty straightforward: select data to plot, go to <strong>Plot</strong> menu, and select <strong>Barchart/Boxplots</strong>. PAST gives you a few options to customize your box plots, most notably the option to enable or disable outliers.</p><p><strong>Pros:</strong> open source (free), comes with various options for plotting and performing statistical analysis.</p><p><strong>Cons:</strong> Only works on Windows; data files need to be formatted correctly to be plottable.</p><h3>3. Excel</h3><p>If you like using Excel and have the 2016 or later versions, we have good news: you can make box and whisker plots in Excel. Unlike earlier versions where you have to do a lot of tweaking to modify column charts into box plots, Excel 2016 and later versions actually come with a chart template for box plot.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/486/0*xZO7HO7GdBg-KU63.png" /></figure><p>Here is how to make a box and whisker plot in Excel: First, select the data to plot. Then go to the <strong>Insert</strong> tab, locate the Chart group, and select the <strong>Statistical Chart</strong> symbol. Click the <strong>Box and Whisker chart</strong> button. Once the box plot is made, you can add a chart title and legend, and customize the plot appearance.</p><p><strong>Pros:</strong> easy to use if you are already familiar with Excel.</p><p><strong>Cons:</strong> limited options to customize plots; only works in Excel 2016 or later.</p><h3>4. BoxPlotR</h3><p>If you want to use R to make box plots but don’t have time to figure out the syntax, <a href="https://matplotlib.org/">BoxPlotR</a> is an excellent solution. BoxPlotR is a web-based application that was written in R, so it has most (if not all) of the customizing options from R, but those options are presented in a very friendly interface (as in: fill in the blanks and you’re done).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*fHNjtqfRQ_NYFlyK.png" /></figure><p>Note that BoxPlotR requires your data file to be formatted in a specific arrangement, that is, identical to their sample dataset, in order to be plotted. There is no option to add legends, so you’ll have to patch those in after downloading the plots. And as the name indicates, BoxPlotR makes box plots and box plots only. If you want a tool that offers a variety of data visualization options, you might have to look elsewhere. Nonetheless, BoxPlotR is still one of the best box and whisker plot makers out there.</p><p><strong>Pros:</strong> customizable; user-friendly; open source.</p><p><strong>Cons:</strong> requires specific data format; only makes box plots.</p><h3>Command line box and whisker plot makers</h3><h3>5. Matplotlib</h3><p>Matplotlib is a Python library — a collection of reusable code — for plotting. Matplotlib is very useful in creating attractive and customizable graphs and plots, as long as you are comfortable with Python.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/0*DPXvQGFp8kB_Tk2c.png" /></figure><p>Matplotlib is a command-line tool and therefore comes with a steep learning curve. You might find it easier to learn if you are already familiar with Matlab, since the Matplotlib interface is similar. Here’s a <a href="http://blog.bharatbhole.com/creating-boxplots-with-matplotlib/">simple tutorial</a> on how to make box plots with Matplotlib.</p><p><strong>Pros:</strong> Creates highly attractive and customizable box plots; open source.</p><p><strong>Cons:</strong> Steep learning curve, only works with Python.</p><h3>6. R</h3><p>R is very popular among researchers who frequently deal with large datasets. Like Matplotlib, its command-line interface is not the most friendly for users who are not familiar with the language. While you might have to spend a considerable amount of time and effort to learn how to plot in R, it does provide an excellent tool for visualizing large datasets.</p><p>You can create box and whisker plots in R using the boxplot() function. Here’s a good tutorial on how to make box plots in R.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*XzEygB67fMZ8_Pxw.png" /></figure><p><strong>Pros:</strong> creates customizable box plots; open source; excellent at handling large datasets.</p><p><strong>Cons:</strong> steep learning curve.</p><p>We hope that the post helps you find the right tool to create your box plots.</p><p>Please leave a comment if you have any questions and share the topic you want us to write in the comment.</p><p>For the original post, <a href="https://blog.bioturing.com/2020/09/18/6-best-box-and-whisker-plot-makers/">visit our blog</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=efdb2acc3952" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Immunoglobulin genes up-regulated in lung adenocarcinoma infiltrating T cells: A report from…]]></title>
            <link>https://bioturing.medium.com/immunoglobulin-genes-up-regulated-in-lung-adenocarcinoma-infiltrating-t-cells-a-report-from-e265643d9634?source=rss-64e2fc4336f3------2</link>
            <guid isPermaLink="false">https://medium.com/p/e265643d9634</guid>
            <category><![CDATA[single-cell-sequencing]]></category>
            <category><![CDATA[single-cell-analysis]]></category>
            <dc:creator><![CDATA[BioTuring Team]]></dc:creator>
            <pubDate>Tue, 08 Sep 2020 12:33:33 GMT</pubDate>
            <atom:updated>2020-12-16T09:39:37.176Z</atom:updated>
            <content:encoded><![CDATA[<h3>Immunoglobulin genes up-regulated in lung adenocarcinoma infiltrating T cells: A report from BioTuring lung cancer single cell database — BioTuring’s Blog</h3><h3>Highlights:</h3><p>In every biology textbook, expression of the B cell receptor (BCR) defines B cells, and the T cell receptor (TCR) defines T cells. However, in 2019, <a href="https://pubmed.ncbi.nlm.nih.gov/31150624/">Ahmed et al.</a> discovered a strange cell population in Type I Diabetes that coexpresses the BCR and TCR, and key lineage markers of both B and T cells.</p><p>In this work, we compared the <strong>T cell populations</strong> in lung tumors vs normal lung tissue from all BioTuring lung cancer single cell data and surprisingly found that <a href="https://en.wikipedia.org/wiki/B-cell_receptor"><strong>B Cell Receptor</strong></a><strong> (BCR) </strong><a href="https://en.wikipedia.org/wiki/Immunoglobulin_light_chain"><strong>light chain</strong></a><strong> and </strong><a href="https://en.wikipedia.org/wiki/Immunoglobulin_heavy_chain"><strong>heavy chain</strong></a> genes (specifically, IGKC and IGHA1) to be up-regulated in lung adenocarcinoma.</p><h3>Comparing T cells populations in lung cancer datasets:</h3><p>Lung cancer is the most common cancer in the world (<a href="https://gco.iarc.fr/today/data/factsheets/populations/900-world-fact-sheets.pdf"><em>Globocan 2018</em></a>) (Fig.1). It can be broadly classified into two forms, small-cell lung carcinomas (SCLCs) and non-small-cell lung carcinomas (NSCLCs). In fact, lung adenocarcinoma is the most popular type of NSCLCs, which also accounts for 30–40% of lung cancer ( <a href="https://pubmed.ncbi.nlm.nih.gov/11944603/">Cersosimo RJ et al., 2002</a>).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*DIiuO8XAjzOiPmQX.png" /></figure><p><strong><em>Figure 1</em></strong><em>. Number of deaths in 2018, both sexes, all ages (Globocan 2018)</em></p><p>To understand the immune cell dynamics in lung adenocarcinoma, we used BioTuring Browser Lung Cancer <a href="https://blog.bioturing.com/2019/06/18/bioturing-browser-making-published-single-cell-data-really-accessible/">Single cell Database</a> (consisting of more than <strong>11.000.000</strong> cells) to explore two lung adenocarcinoma datasets:</p><p>(1) Single-cell RNA sequencing demonstrates the molecular and cellular reprogramming of metastatic lung adenocarcinoma <em>(</em><a href="https://www.nature.com/articles/s41467-020-16164-1"><strong><em>Kim et al., 2020</em></strong></a><em>)</em>.</p><p>(2) Regenerative lineages and immune-mediated pruning in lung cancer metastasis <em>(</em><a href="https://www.nature.com/articles/s41591-019-0750-6"><strong><em>Laughney et al., 2020</em></strong></a><em>).</em></p><h3>1.IGKC, IGLC2, IGHA1, and IGHG3 are up-regulated in T cells from lung tumor and brain metastasis in the dataset of Kim et al. (2020)</h3><p>Using BioTuring Browser to find differentially expressed genes in T cells from the tumor vs. normal lung tissues, we found significant up-regulation of immunoglobulin genes in tumor-infiltrating T cells. These most significant genes included IGKC, IGLC2, IGHA1 and IGHG3 (Fig. 3, 4).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*HDIfKQAUK9TNCNWq.png" /></figure><p><strong><em>Figure 2.</em></strong><em>(A) t-SNE of all cell types (left) and (B) only the T cells (right) in the dataset of Kim et al. (2020), colored by normal lung (42.995 cells) and tumor lung (45.149 cells)</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/893/0*gOhhuPeqqX3DlxDW.png" /></figure><p><strong><em>Figure 3.</em></strong><em> Heatmap of top 10 differentially expressed genes in T cells from the lung tumor vs. normal lung tissue</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/855/0*T3_sRtdj48PcGWn8.png" /></figure><p><strong><em>Figure 4. </em></strong><em>Top 4 Immunoglobulin genes up-regulated in Tumor T cells from left to right: IGKC, IGLC2, IGHG3 and IGHA1. Upper panel shows T cells from lung cancer while the lower panel shows T cells from normal lungs.</em></p><p>Meanwhile, IGKC, the most significant gene, was also highly up-regulated in the T cells of brain metastasis, compared with those in the normal lung tissues (Fig. 3). Our composition analysis for T-cells in tumors also showed that <a href="https://www.novusbio.com/research-areas/immunology/adaptive-immunity"><em>CD8+/CD4+ mixed T helper cells</em> </a>were the subtypes that exhibited the highest expression of IGKC (Fig. 5).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*N71nnlCBF3JieqzH.png" /></figure><p><strong><em>Figure 5. </em></strong><em>(A) t-SNE of T cells from 3 groups: Brain Metastasis, Normal (Lung) and Tumor (lung), (B) IGKC expression level in T cells, (C ) IGKC expression across T cells from Brain Metastasis, Normal (Lung) and Tumor (lung), (D) T cell subtypes identified in the study, (E) Composition analysis showed that top IGKC expressors in Tumor (lung) were mainly CD8+/CD4+ mixed T helper cells</em></p><h3>2.IGKC, IGLC2, IGLC3 and IGHA1 expression gradually increased in T cells from normal tissues, to tumor, to metastasis in the dataset of Laughney et al. (2020)</h3><p>In the dataset of Laughney and colleagues (2020), we also ran differential expression analysis on CD3+ cell populations from the tumor vs. normal lung tissue. And immunoglobulin genes is also up-regulated in the T cells from tumor and metastasis groups, compared with normal lung tissues (Fig. 6 A,B,C).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/886/0*SWeqkxJ-aYVU13-y.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/886/0*22CPmJ5ImeBbqdTz.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/887/0*Wlnhq1A5t7_fDcXr.png" /></figure><p><strong><em>Figure 6.</em></strong><em> Differential expression analysis between T cell population of (A) Tumor vs. Normal groups, (B) Metastasis vs. Tumor, (C )Metastasis vs. Normal</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*kFBR6dpHpWFwCtP2.png" /></figure><p><strong><em>Figure 7.</em></strong><em> (A) t-SNE of the T cell population of 3 groups: Metastasis, Normal and Tumor, and (B) all T cell subtypes in 3 groups; (C ) t-SNE of IGKC expression and (D) comparison of IGKC expression level across 3 conditions in the entire T cell population and in (E) each T cell subtype.</em></p><p>Interestingly, in this dataset, IGKC was gradually up-regulated in T cells from normal tissues to tumor to metastasis. In particular, this appeared in the memory T cell, NK T cell, and T helper cell (Fig. 7).</p><h3>Discussion</h3><p><a href="https://pubmed.ncbi.nlm.nih.gov/31150624/">In 2019, Ahmed et al.</a> discovered a dual expresser cell population in Type 1 Diabetes which expresses genes specific to both B and T cell lineages and also their receptors. These dual expresser cells are marked by <a href="https://en.wikipedia.org/wiki/B-cell_receptor">BCR</a> (CD79a, CD79b)/(CD19 and CD20) with light/heavy chains and <a href="https://en.wikipedia.org/wiki/T-cell_receptor">TCR</a> (CD3G, CD3E, CD3D, and CD247) (Ahmed et al.,2019).</p><p>What is the role of these dual expresser cells? In type I diabetes, HLA-DQ8 is known to mistakenly present insulin to T cells in such a way that stimulates T cells to attack the pancreatic β cells, which produce insulin. Interestingly, these dual expresser cells have a <a href="https://ars.els-cdn.com/content/image/1-s2.0-S0092867419305057-fx1.jpg">single BCR clonotype</a> (IGHV), which produces a peptide that is bound tightly to the HLA-DQ8 molecule. This stimulated T helper cells from Type 1 Diabete patients to proliferate and to secrete proinflammatory cytokines. The T helper cells then go on to direct certain killer T cells to eliminate β cells of the pancreas ( <a href="https://pubmed.ncbi.nlm.nih.gov/31150624/">Ahmed et al</a>., 2019).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*mmn8l4F-rqWIsHdd.png" /></figure><p><em>Figure 8. Dual expresser (DE) cells comprise of both</em> <em>BCRs and TCRs in T1D (Ahmed et al.,2019)</em></p><p><strong><em>Does it mean the immunoglobulin up-regulation in tumor-infiltrating T cells triggers the immune response? And against what?</em></strong></p><p>To answer this, we need more data, and research across other cancer types, to understand whether this immunoglobulin up-regulation also happens consistently.</p><p>For detailed instructions on how to reproduce the analysis, check out our video below:</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2F_u9hGI-ofao%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3D_u9hGI-ofao&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2F_u9hGI-ofao%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/a5878f2c6c033c648769e6195c40265c/href">https://medium.com/media/a5878f2c6c033c648769e6195c40265c/href</a></iframe><p><strong><em>This article is for research purposes only. Please do not use this for clinical purposes to any extent.</em></strong></p><h3>References</h3><ol><li>Cersosimo RJ. Lung cancer: a review. <em>Am J Health Syst Pharm</em>. 2002;59(7):611–642. doi:<a href="https://doi.org/10.1093/ajhp/59.7.611">10.1093/ajhp/59.7.611</a></li><li>Neeve SC, Robinson BW, Fear VS. The role and therapeutic implications of T cells in cancer of the lung. <em>Clin Transl Immunology</em>. 2019;8(8):e1076. Published 2019 Aug 28. doi:<a href="https://onlinelibrary.wiley.com/doi/full/10.1002/cti2.1076">10.1002/cti2.1076</a></li><li>Rizwan Ahmed et al. A Public BCR Present in a Unique Dual-Receptor-Expressing Lymphocyte from Type 1 Diabetes Patients Encodes a Potent T Cell Autoantigen, 2019, Cell. Volume 177, Issue 6,1583–1599.e16. <a href="https://doi.org/10.1016/j.cell.2019.05.007">https://doi.org/10.1016/j.cell.2019.05.007</a></li><li>Nayoung Kim et al. Single-cell RNA sequencing demonstrates the molecular and cellular reprogramming of metastatic lung adenocarcinoma. 2020. Nat Commun <strong>11, </strong>2285 (2020). <a href="https://doi.org/10.1038/s41467-020-16164-1">https://doi.org/10.1038/s41467-020-16164-1</a></li><li>Laughney A.M., Hu, J., Campbell, N.R. <em>et al.</em> Regenerative lineages and immune-mediated pruning in lung cancer metastasis. 2020. <em>Nat Med</em> <strong>26, </strong>259–269. <a href="https://doi.org/10.1038/s41591-019-0750-6">https://doi.org/10.1038/s41591-019-0750-6</a></li><li>Lehuen, A., Diana, J., Zaccone, P. <em>et al.</em> Immune cell crosstalk in type 1 diabetes. 2010. <em>Nat Rev Immunol</em> <strong>10, </strong>501–513. <a href="https://doi.org/10.1038/nri2787">https://doi.org/10.1038/nri2787</a></li></ol><p><em>Originally published at </em><a href="https://blog.bioturing.com/2020/09/08/immunoglobulin-genes-in-lung-adenocarcinoma-t-cells-a-report-from-bioturing-lung-cancer-single-cell-rna-seq-database/"><em>https://blog.bioturing.com</em></a><em> on September 8, 2020.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e265643d9634" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Single-cell analysis of CD8+ T cells in immune checkpoint blockade: some reproducible insights…]]></title>
            <link>https://bioturing.medium.com/single-cell-analysis-of-cd8-t-cells-in-immune-checkpoint-blockade-some-reproducible-insights-54db9aa8b111?source=rss-64e2fc4336f3------2</link>
            <guid isPermaLink="false">https://medium.com/p/54db9aa8b111</guid>
            <category><![CDATA[single-cell-sequencing]]></category>
            <category><![CDATA[single-cell-analysis]]></category>
            <dc:creator><![CDATA[BioTuring Team]]></dc:creator>
            <pubDate>Mon, 30 Mar 2020 05:04:38 GMT</pubDate>
            <atom:updated>2020-12-15T09:19:28.668Z</atom:updated>
            <content:encoded><![CDATA[<h3>Single-cell analysis of CD8+ T cells in immune checkpoint blockade: some reproducible insights from BioTuring Database — BioTuring’s Blog</h3><p>Cancer treatment, as challenging as it sounds, has been revolutionized by recent advances in immune checkpoint blockade (ICB) therapies, which use monoclonal antibodies (mAb) to block the suppressive immune checkpoints, helping the patient’s immune system to kill cancer cells. However, despite several clinical case studies of success, some patients still do not respond well or not at all to these therapies.</p><p><strong>What could be an explanation to these negative responses?</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*QdFCUlw8-cRlllyu.png" /></figure><p><strong><em>Figure 1: Illustration of PD-1 checkpoint blockade.</em></strong></p><p>One promising way to answer the question is to fully characterize the heterogeneity and dynamics of tumor infiltrating lymphocytes (TILs) before, during, and after ICB for cancer treatment. Thankfully, with recent advances in single-cell sequencing technologies, the landscape of tumor TILs in immunotherapy has been gradually completed at an unprecedented resolution, shedding light on the immune system dynamics and its relationship with clinical responses in specific groups of patients (<em>Tirosh et. al., 2016; Clark et. al., 2019; Yost et. al., 2019, etc.</em>)</p><p>In this work, we use multiple single-cell transcriptome data sets from <a href="https://blog.bioturing.com/2019/06/18/bioturing-browser-making-published-single-cell-data-really-accessible/"><strong>BioTuring Browser</strong></a> to highlight shared perspectives on using the TIL landscape for predicting response to immunotherapy. We focus, in particular, on CD8+ T cells, a very complex and dynamic population that has shown to impact clinical responses in specific patient groups.</p><h3>Enrichment of exhausted CD8+ T cells in non-respondents</h3><p>The trajectory begins with regulatory T cells, memory T cells, lymphocytes, to exhausted CD8+ T cells that express heat shock proteins, exhausted CD8+, and exhausted T cells that express cell-cycle genes. Interestingly, most cells from responding patients are positioned near the upper bound, while non-responder samples stretch substantially to exhausted/dysfunctional states (<strong>Fig. 2a,b,d</strong>), and also form a separate cluster (marked by some TRD genes, KLRG1) ( <strong>Fig. 2c</strong>).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*JbGcJu2uY9dANhsV.png" /></figure><p><strong><em>Figure 2. Sub-clustering of T cells in the dataset of Sade-Feldman &amp; Yizhak et. al. (2018).</em></strong> <strong><em>(a)</em></strong><em> UMAP of all T cell sub-clusters shows a transition of T cells from memory to exhausted states. The trajectory begins with regulatory T cells, memory T cells, lymphocytes, to exhausted CD8+ T cells that express heat shock proteins, exhausted CD8+, and exhausted T cells that express cell-cycle genes; </em><strong><em>(b)</em></strong><em> T cells colored by responder and non-responder; </em><strong><em>(c ), (d)</em></strong><em> Log2 (raw expression values) of TRDC and LAG3 (an exhaustion marker)</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/795/0*zioU2fh6hiZe8AhM.png" /></figure><p><strong><em>Figure 3. Cell type composition of (a) responders and (b) non-responders.</em></strong></p><h3>TCF7+ PD-1- CD8+ T cell abundance: a predictive marker for checkpoint blockade outcomes</h3><p>Sade-Feldman &amp; Yizhak et. al. also pinpointed the transcription factor 7 (TCF7) expression in CD8+ T cells as a <em>predictive </em>marker for immune checkpoint blockade responses.</p><p>To find whether this finding is reproducible, we followed the authors’ study to select and sub-cluster the CD8+ T cell populations to just focus on its subsets and dynamics. To select these cells, we used a bi-axial feature plot, graphing cells by CD3+ and CD8+ expression <strong>(Fig. 4a) </strong>and select the cluster in the middle, containing cells that expressed both genes. After sub-clustering and coloring the CD8+ T cells by responder and non-responder, we identified a special CD8+ T cell subpopulation highly specific to responders ( <strong>Fig. 4b</strong>). We then proceeded to find the marker genes for this population.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/897/0*ir38VXyfGR_fbySM.png" /></figure><p><strong><em>Figure 4. TCF7+ population enriched in respondents pre-treatment. (a).</em></strong><em> Feature plot of CD3D and CD8 expression; </em><strong><em>(b)</em></strong><em> t-SNE of CD8+ CD3+ cells colored by response and non-response, showing a CD8+ T cell subpopulation highly specific to respondents; </em><strong><em>(c )</em></strong><em> list of marker genes for the response-specific CD8+ sub-population; </em><strong><em>(d)</em></strong><em> TCF7 expression across all CD8+ T cells; </em><strong><em>(e) </em></strong><em>violin plot of TCF7 expression between response and non-response; </em><strong><em>(f) </em></strong><em>composition of the TCF7+ response-specific population before treatment of respondents and non-respondents. The percentage of TCF7+PD1- is higher in the response group.</em></p><p>Notably, TCF7, which is part of the Wnt/ b-catenin signaling pathway, was previously known to reinvigorate CD8+ T cells in response to PD-1 blockade to resolve chronic infection (Im et al., 2016; Utzschneider et al., 2016). Zhou et al. (2010) also suggested that the gene was crucial for differentiation, self-renewal, and persistence of memory CD8+ T cells, which could potentially explain its link to favourable clinical responses of ICB in melanoma.</p><h3>Is TOX a useful predictor for PD-1 blockade?</h3><p>TOX (Thymocyte Selection Associated High Mobility Group Box) has been shown to promote T cell exhaustion (Khan et. al., 2019; Seo et. al., 2019). TOX could impair CD8+ T cell anti-tumor function and response to anti-PD1 therapy (Wang et. al., 2019). A recent study by Kim et. al. (2020) also confirmed that the expression levels of TOX in tumor infiltrating T cells could predict response to anti-PD-1 therapy in human melanoma and non-small-cell lung cancer, and that TOX inhibition may potentially impede T cell exhaustion and improve ICB efficacy.</p><p>We hypothesized that TOX could also link to T cell exhaustion. We verified this assumption using multiple single-cell data sets in BioTuring database. The question we had was how to find single-cell data sets with TOX genes expressed. Fortunately, BioTuring API allows us to perform such a query.</p><p>Using BioTuring API, we queried the TOX gene across all cancer related studies. Interestingly, BioTuring database returned exhausted T cells in 4 data sets: Zhang et. al (2018), Sade-Feldman et. al. (2018), Guo et. al. (2019) and Yost et. al. (2019) (Fig 5), where TOX up-regulated in these exhausted T-cell populations.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*RObZlnwySQgBnMWa.png" /></figure><p><strong><em>Figure 5: High TOX expression in T follicular helper cell (Tfh) and exhausted T cell populations in the single-cell dataset of Yost et. al. (2019).</em></strong></p><p>Next, we used the dataset of Sade-Feldman to study TOX expression in CD8+ T cells of respondents vs. non-respondents, before and after ICB. Interestingly, we noticed an increase of expression in non-respondents before treatment (p=3.13e-07) (Fig 6). TOX-positive cells shared 59.14% of the total CD8+ T cells in non-respondents, while accounting for 42.86% in the response group. Altogether, our reanalysis of the database confirms TOX’s link to T cell exhaustion, and its potential role as a predictive marker for non-response</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/300/0*QcmPtro1ePa8DkUL.png" /></figure><p><strong><em>Fig 6: Student’s t-test of TOX expression between non-respondents and respondents before ICB (performed with </em></strong><a href="https://vinci.bioturing.com/"><strong><em>BioVinci</em></strong></a><strong><em>).</em></strong></p><p>Above are some reproducible insights into the TIL dynamics under ICB, reanalyzed using BioTuring’s public single-cell data repository. We hope this helps to confirm the findings in a data-centric approach for other studies such as this one, and at the same time provides more insights into the CD8+ T cell dynamics under immune checkpoint blockade. Further analyses into phenotype-clonotype stability and clonal replacement of exhausted CD8+ T cells following PD-1 blockade will soon be coming in the next episode. These insights are really important to better understand the dynamics in immuno-oncology therapies and to understand why groups of patients are better responders than others.</p><p><strong>References</strong></p><p>Clarke, James, et al. “Single-cell transcriptomic analysis of tissue-resident memory T cells in human lung cancer.” Journal of Experimental Medicine 216.9 (2019): 2128–2149.</p><p>Guo, Xinyi, et al. “Global characterization of T cells in non-small-cell lung cancer by single-cell sequencing.” Nature medicine 24.7 (2018): 978–985.</p><p>Im, Se Jin, et al. “Defining CD8+ T cells that provide the proliferative burst after PD-1 therapy.” Nature 537.7620 (2016): 417–421.</p><p>Khan, Omar, et al. “TOX transcriptionally and epigenetically programs CD8+ T cell exhaustion.” Nature 571.7764 (2019): 211–218.</p><p>Kim, Kyungsoo, et al. “Single-cell transcriptome analysis reveals TOX as a promoting factor for T cell exhaustion and a predictor for anti-PD-1 responses in human cancer.” Genome Medicine 12.1 (2020): 1–16.</p><p>Kurtulus, S. et al. Checkpoint blockade immunotherapy induces dynamic changes in PD-1−CD8+ tumor infiltrating T cells. Immunity 50, 181–194.e6 (2019).</p><p>Miller, B. C. et al. Subsets of exhausted CD8+ T cells differentially mediate tumor control and respond to checkpoint blockade. Nat. Immunol. 20, 326–336 (2019).</p><p>Siddiqui, I. et al. Intratumoral Tcf1+PD-1+CD8+ T cells with stem-like properties promote tumor control in response to vaccination and checkpoint blockade immunotherapy. Immunity 50, 195–211.e10 (2019).</p><p>Tirosh, Itay, et al. “Dissecting the multicellular ecosystem of metastatic melanoma by single-cell RNA-seq.” Science 352.6282 (2016): 189–196.</p><p>Seo, Hyungseok, et al. “TOX and TOX2 transcription factors cooperate with NR4A transcription factors to impose CD8+ T cell exhaustion.” Proceedings of the National Academy of Sciences 116.25 (2019): 12410–12415.</p><p>Utzschneider, Daniel T., et al. “T cell factor 1-expressing memory-like CD8+ T cells sustain the immune response to chronic viral infections.” Immunity 45.2 (2016): 415–427.</p><p>Wang, Xiaochen, et al. “TOX promotes the exhaustion of antitumor CD8+ T cells by preventing PD1 degradation in hepatocellular carcinoma.” Journal of hepatology 71.4 (2019): 731–741.</p><p>Yost, Kathryn E., et al. “Clonal replacement of tumor-specific T cells following PD-1 blockade.” Nature medicine 25.8 (2019): 1251–1259.</p><p>Zhou, Xinyuan, et al. “Differentiation and persistence of memory CD8+ T cells depend on T cell factor 1.” Immunity 33.2 (2010): 229–240.</p><p><em>Originally published at </em><a href="https://blog.bioturing.com/2020/03/30/single-cell-analysis-of-cd8-t-cells-in-immune-checkpoint-blockade-some-reproducible-insights-from-bioturing-database/"><em>https://blog.bioturing.com</em></a><em> on March 30, 2020.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=54db9aa8b111" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A review of Haining Lab’s work: Loss of ADAR1 in tumors overcomes resistance to immune checkpoint…]]></title>
            <link>https://bioturing.medium.com/a-review-of-haining-labs-work-loss-of-adar1-in-tumors-overcomes-resistance-to-immune-checkpoint-22f76f802223?source=rss-64e2fc4336f3------2</link>
            <guid isPermaLink="false">https://medium.com/p/22f76f802223</guid>
            <category><![CDATA[single-cell-sequencing]]></category>
            <category><![CDATA[single-cell-analysis]]></category>
            <category><![CDATA[immunotherapy]]></category>
            <dc:creator><![CDATA[BioTuring Team]]></dc:creator>
            <pubDate>Fri, 04 Jan 2019 06:38:18 GMT</pubDate>
            <atom:updated>2020-12-15T09:07:49.237Z</atom:updated>
            <content:encoded><![CDATA[<h3>A review of Haining Lab’s work: Loss of ADAR1 in tumors overcomes resistance to immune checkpoint blockade — BioTuring’s Blog</h3><p>Ever questioned why the immune system does not attack our own double-stranded RNA? Speaking of this, we should not ignore the role of ADAR, the gene that encodes Double-stranded RNA-specific adenosine deaminase enzyme, responsible for converting adenosines to inosines (A -&gt; I editing) in double-stranded RNA (dsRNA) substrates. ADAR was found to be involved in the sensing pathway to endogenous double-stranded RNA by Mannion and colleagues in 2014. In that study, Adar1 mutant mouse embryos showed aberrant interferon induction and died by embryonic day E12.5. ADAR is great for maintaining our being, isn’t it?</p><p>Not just for our being, but potentially for the “tumor being” as well?</p><p>A couple of weeks ago, Ishizuka and colleagues from Haining Lab published their work on the role of ADAR1 in immunotherapy resistance, showing its loss profoundly sensitized tumors to treatment. Using <em>single-cell RNA sequencing </em>on CD45+ immune cells from Adar1-null (Adar1-knockout) and control tumors [and other techniques], the authors were able to compare the immune microenvironment in two models, and detect a significant decrease in the proportion of myeloid-derived suppressor cells (known for suppressing T cell responses) and M2 macrophages (known for tumor promotion) in Adar1-knockout. Interestingly, up-regulated genes are highly enriched in the biological process “ <em>respond to type I interferon </em>“.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/685/0*8OmYbvQ-Z53jWhcP.jpg" /></figure><p><em>Source: Ishizuka et al. 2018</em></p><p>Such an amazing work gained so much interest from the scientific community. We have received a lot of requests to index the single-cell RNA-seq dataset of this paper into BBrowser. Now the data are ready. Below, we provide details on how we reproduce findings in this paper using <a href="https://blog.bioturing.com/2018/12/05/bioturing-browser-the-software-to-resolve-major-challenges-in-single-cell-rna-seq-data-analysis/">BBrowser</a>. This includes the text guide as well as a screencast video.</p><h3>1. The experimental design</h3><p>In a previous work from Haining lab, Manguso et al., 2017 identified Adar1 as one of the genes with a potential to modify the response to endogenous RNA species.</p><p>This time, Ishizuka and colleagues showed Adar1 functions as a checkpoint that limits anti-tumor immunity by preventing the sensing of endogenous dsRNA. This paper performed <em>single-cell RNA sequencing on CD45+ immune cells </em>from Adar1-null and control tumors (implanted into C57BL/6 female mice) — to compare the immune microenvironment of these two models. The authors also added B16 tumor cells grown in culture to each sample at a 5% ratio as a spike-in control for assessing sample-to-sample variability. The total 8834 cells (including both infiltrating immune cells and spiked-in tumor cells) were counted and loaded into the 10xGenomics device.</p><h3>2. Marking the spiked-in tumor cells</h3><p>As the main interest of the study is the infiltrating immune cell population, we locate and exclude the spiked-in tumor cells in the data.</p><p>We, therefore, choose to see the expression of both and (markers for tumor cells) in the cell population. Cells with high Pmel expression are in red, while cells with high Mlana expression are in blue. Cells that express both genes are marked in yellow. The yellow cluster on the left of the t-SNE plot corresponds to the spiked-in tumor cells.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*XNLpNPGNG6Hwou7d.png" /></figure><p><em>Fig.1: Yellow clusters: spiked-in tumor cells</em></p><h3>3. Proportions of MDSCs and M2 macrophages decrease in Adar1-null</h3><p>In their study, Ishizuka and colleagues noticed a decreased proportion of MDSCs (myeloid-derived suppressor cells — known for the ability to T cell responses) and M2 macrophage populations (known for tumor promotion) in Adar1-null model. Let’s check this out by BBrowser within a few steps.</p><p>As there are two replicates for each condition (Adar1-null and Control), we will see four groups in the metadata: ADAR_S1, ADAR_S2, Control_S3, Control_S4. We merge ADAR_S1 and ADAR_S2 into “Adar1-null”. We merge Control_S3 and Control_S4 into Control.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*uluHa-xf48LZbzvD.png" /></figure><p><em>Fig. 2a: Merging ADAR_S1 and ADAR_S2 into “Adar1-null” group.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*fmxncONGMZkj7Gbc.png" /></figure><p><em>Fig. 2b: The finished plot that merged replicates into one group</em></p><p>After merging the metadata, we identify MDSC population using marker gene Cxcl3. We get the exact number of cells with positive expression in each group: Control (442), Adar1-null (80), and Spiked-in tumor cells (87). The remaining job is to find and compare the ratios of Cxcl3+ cells in Adar1-null and Control. This ratio in Adar1-null is 80/3298=2.4%, which is extremely low compared to the ratio in Control: 442/4216=10.4%.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*i79L0KEo3iNB0y7l.jpg" /></figure><p><em>Fig. 3: M2 macrophage cluster (left: Sepp1, right: Clec10a)</em></p><p>We detect M2 macrophage population using Sepp1 and Clec10a as marker genes (Fig. 3). Again, we want to see the differences of M2 macrophage population in the two conditions: Adar1-null and Control. As shown in BBrowser, 86% of the cells in the M2 macrophage cluster are originated from Control mice.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Sso-8VdEFHG22q5n.png" /></figure><p><em>Fig. 4: M2 macrophage composition in the two conditions: Adar1-null and Control</em></p><h3>4. Comparing Adar1-null and Control in infiltrating immune cells</h3><p>“Deletion of Adar1 causes a global reshaping of the tumor immune compartment and increased abundance of IFNs.”</p><p>To produce this in BBrowser, we perform Gene Set Enrichment Analysis (GSEA) for the most up-regulated genes in Adar1-null compared with the Control model. As a result, “response to type I interferon” was identified in the top list of gene ontologies.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ZR65VdFUSLNerOZ7.png" /></figure><p><em>Fig. 5 Gene Set Enrichment Analysis results for up-regulated genes in Adar1-null compared with Control.</em></p><p>Above is our recording of how we explored this amazing work. Apart from these analyses, there are many more functions that you can perform using BBrowser (eg. sub-clustering, DE analysis, and real-time cell-type prediction), and have many flexibilities to explore this dataset in your own way. We welcome all the valuable comments on BBrowser. It is now available for download at <a href="http://www.bioturing.com/product/bbrowser">www.bioturing.com/product/bbrowser </a>.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FJvoIS2yqFME%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DJvoIS2yqFME&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FJvoIS2yqFME%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/1f035cde5b05b47e698dd24c15bfe0e1/href">https://medium.com/media/1f035cde5b05b47e698dd24c15bfe0e1/href</a></iframe><p><strong>References</strong></p><ol><li>Manguso, R. T. In vivo CRISPR screening identifies Ptpn2 as a cancer immunotherapy target. Nature 547, 413–418 (2017).</li><li>Mannion, N. M. The RNA-editing enzyme ADAR1 controls innate immune responses to RNA. Cell Rep. 9, 1482–1494 (2014).</li><li>Ishizuka, J. J. Loss of ADAR1 in tumours overcomes resistance to immune checkpoint blockade. Nature 565, 43–48 (2019).</li><li>Srivastava, M.K., Sinha, P., Clements, V.K., Rodriguez, P. &amp; Ostrand-Rosenberg, S. Myeloid-derived suppressor cells inhibit T-cell activation by depleting cystine and cysteine. Cancer Res. 70, 68–77 (2010).</li><li>Yuan A. et al. Opposite effects of M1 and M2 macrophage subtypes on lung cancer progression. Sci. Rep. 5, 14273 (2015)</li></ol><p><em>Originally published at </em><a href="https://blog.bioturing.com/2019/01/04/loss-of-adar1-in-tumours-overcomes-resistance-to-immune-checkpoint-blockade/"><em>https://blog.bioturing.com</em></a><em> on January 4, 2019.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=22f76f802223" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[BioTuring Browser: the software to resolve major challenges in single-cell RNA-seq data analysis —…]]></title>
            <link>https://bioturing.medium.com/bioturing-browser-the-software-to-resolve-major-challenges-in-single-cell-rna-seq-data-analysis-8d0ba7ef894f?source=rss-64e2fc4336f3------2</link>
            <guid isPermaLink="false">https://medium.com/p/8d0ba7ef894f</guid>
            <dc:creator><![CDATA[BioTuring Team]]></dc:creator>
            <pubDate>Wed, 05 Dec 2018 06:38:16 GMT</pubDate>
            <atom:updated>2020-12-15T08:58:01.936Z</atom:updated>
            <content:encoded><![CDATA[<h3>BioTuring Browser: the software to resolve major challenges in single-cell RNA-seq data analysis — BioTuring’s Blog</h3><p>Single-cell RNA-seq technologies have opened up a completely new era for transcriptomic studies. For the first time ever, scientists can look at individual transcriptomic profiles of millions of cells, and better understand how each cell functions in a tissue. Yet science is confronting bigger challenges analyzing these massive amounts of data from single-cell RNA-seq: finding the best algorithms to quickly and accurately process such data, and a visualization platform that is strong enough to interactively graph hundreds of thousands to millions of data points.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/0*_RWkJdaiqpBZvR7t.gif" /><figcaption>Mouse Organogenesis Cell Atlas (Cao et.al., 2019) in BioTuring Browser</figcaption></figure><p>We introduce <a href="https://bioturing.com/product/bbrowser"><strong>BioTuring Browser</strong></a>, the software to tackle major challenges in single-cell RNA-seq data analysis that is extremely easy to use for any biomedical scientists. BioTuring Browser dashboards are interactive, compact enough to run on a standard laptop, and most importantly, has the power to process and visualize up to 1.3 million cells at a time.</p><p>BioTuring Browser is the first software to provide 3D rotation, differential expression analysis, and batch effect removal for many batches of scRNA-seq data. With the help of machine learning and AI, the software can run real-time cell type prediction, helping researchers and biomedical scientists quickly annotate cell populations. At the moment, our cell type prediction algorithm is optimized for neural cells and immune cells — the final aim is to cover all cell types in the body.</p><p>Here’s to name a few analyses integrated in BioTuring Browser:</p><ul><li>Transcript quantification with high accuracy and speed (using <a href="https://blog.bioturing.com/2018/10/12/cell-ranger-problems-and-hera-t/">Hera-T</a>)</li><li>Integrate Seurat and Scanpy objects to the pipeline</li><li>Batch effect removal with CCA, MNN, and Harmony correction</li><li>Interactive 2- and 3-D visualization of up to 1.3 million cells</li><li>Marker gene detection and real-time cell type prediction</li><li>Sub-clustering</li><li>Differential expression analysis</li><li>Pseudotime lineage reconstruction</li><li>Pairing clonotype data with single-cell expression data</li><li>Analyzing cell surface data from CITE-Seq/ 10X Genomics Feature Barcoding technology</li></ul><p>Together with a full range of analyses and visualizations, BioTuring Browser also provides scientists with <em>easy access </em>to latest scRNA-seq datasets from published studies, making it extremely easy to view and reanalyze valuable past works and compare such with their current studies. Above all, this library of published datasets promises to remove any barriers to access published data and make science completely transparent.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FJvoIS2yqFME%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DJvoIS2yqFME&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FJvoIS2yqFME%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/05707a7fe4b960bd40fb9af770425a9a/href">https://medium.com/media/05707a7fe4b960bd40fb9af770425a9a/href</a></iframe><p>BioTuring Browser is now available for Windows, Mac, and Linux at <a href="http://www.bioturing.com/product/bbrowser">www.bioturing.com/product/bbrowser </a>. We also look forward to comments from users to improve the software, and if you need to adapt and customize BioTuring Browser for your company/organization, please don’t hesitate to contact us at: <a href="/service/mailto:support@bioturing.com">support@bioturing.com</a></p><p><em>Originally published at </em><a href="https://blog.bioturing.com/2018/12/05/bioturing-browser-the-software-to-resolve-major-challenges-in-single-cell-rna-seq-data-analysis/"><em>https://blog.bioturing.com</em></a><em> on December 5, 2018.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8d0ba7ef894f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Dos and don’ts for a heatmap color scale]]></title>
            <link>https://bioturing.medium.com/dos-and-donts-for-a-heatmap-color-scale-75929663988b?source=rss-64e2fc4336f3------2</link>
            <guid isPermaLink="false">https://medium.com/p/75929663988b</guid>
            <category><![CDATA[heatmap]]></category>
            <category><![CDATA[heat-map-color-scale]]></category>
            <category><![CDATA[hierarchical-clustering]]></category>
            <category><![CDATA[heatmap-color-scale]]></category>
            <category><![CDATA[biovinci]]></category>
            <dc:creator><![CDATA[BioTuring Team]]></dc:creator>
            <pubDate>Mon, 24 Sep 2018 03:32:45 GMT</pubDate>
            <atom:updated>2018-10-03T04:57:47.788Z</atom:updated>
            <content:encoded><![CDATA[<p>Heatmaps. We see them all over the scientific journals — the shading matrices that convey meaningful stories. Wanna interpret gene expressions? Draw a heatmap. Gotta examine the correlation among variables? Get it done with a heatmap. It seems that the shading has breathed such life into those cheerless matrices of plain numbers, making it really easy to compare and spot the differences among groups of interest.</p><p>Yet, like anything else, the heatmap color scale can backfire when mistakenly chosen. Either it causes confusion or dizziness, or “heat” of annoyance to your readers. We really do not want to turn a heatmap into a literal “heat” map in this way, do we?</p><p>Here’s a couple of dos and don’ts for a heatmap color scale.</p><h3>Do #1: Use the right kind of color scale</h3><p>When it comes to heatmaps, the two most common ranges of colors are sequential and diverging scales.</p><p><em>Sequential scales</em> use the blended progression, typically of a single hue, from the least to the most opaque shades, representing low to high values (an example is the ColorBrewer Blues scale). Scales that use multiple hues are also considered sequential when hues progress in a single direction from one end to another (like the Viridis scale).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/300/0*uHGU-gdSfjkh4jyp" /><figcaption>Popular sequential heatmap color scales</figcaption></figure><p><em>Diverging scales</em>, on the other hand, show color progression in two directions: gradually toning down the first hue from one end to a neutral color at the midpoint, then increasing the opacity of the second hue to the other end of the scale.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/300/0*UY7SrzuSU6EUVOwn" /></figure><p>Just go for the sequential scale when you need to differentiate high values from low values. When a <em>reference</em> value is in the middle of the data range (such as zero or an average value), you should use a diverging color scale, with a neutral color representing the reference value. Let’s say you want to build a heatmap of gene expression. A sequential color scale is ideal for showing raw TPM values (all of which are non-negative), while a diverging scale will effectively show standardized TPM values (including those of up-regulated and down-regulated genes).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/300/0*uMNEgIRG2rKo3Pha" /><figcaption><em>A sequential scale is good for showing raw TPM values. Got a standardized TPM matrix? Go for the diverging scale.</em></figcaption></figure><h3>Do #2: Find a color-blind-friendly combination</h3><p>Five percent of the entire population will thank you! Why limit people from seeing your meaningful visualizations? Color-blind people tend to struggle with the following combos, depending on their conditions: red-green, green-brown, green-blue, blue-gray, blue-purple, green-gray, green-black and light green-yellow. So try to avoid such combos and go for a color-blind-friendly heatmap palette.</p><p>Here are a couple of combos that work for a heatmap color scale:</p><ul><li>blue &amp; orange,</li><li>blue &amp; red,</li><li>blue &amp; brown</li></ul><p>One more: color-blind people can detect the contrast and opacity, regardless of their impairment. Keep that in mind. There are a lot of options available, right?</p><h3>Don’t #1: Use the “rainbow” scale</h3><p>The “rainbow” scale has had its presence on many scientific papers. No doubt that it’s more attractive than a single-hue color scale. But it can be misleading.</p><p>One of the problems with the rainbow scale is that people automatically see the brightest color as a peak. Some of your readers might see yellow as the highest value. Another one associates that with orange, or blue. There’s no clear and consistent direction in such a scale.</p><p>The rainbow scale also creates the misperception of the magnitude of the data. The values change smoothly, but the colors do not. Abrupt changes between different hues (green to yellow or blue to green) make the values seem significantly distant, while they are actually 0.1–0.2 units far from each other.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/901/0*D1vkL-OeUThy7T8Z" /></figure><p><em>A rainbow color scale is not a good idea for visualizing your data.</em></p><h3>Don’t #2: Get greedy with the colors</h3><p>Don’t put too much color into your heatmap palette. Keep it simple and interpretable. Sequential scales using multiple hues are good to go if the hues closely relate to each other. Otherwise, the heatmap might end up looking exactly like a colorful mosaic that seems attractive at first, but leaves the readers with a dozen of questions. If you once stumble on a green-blue-purple-red scale, you will have to stop and think which two colors are at the extremes. The best option is to pick out 3 consecutive hues on a basic color wheel.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/901/0*CKui841ymfrxzAUo" /><figcaption><em>A simple color scale of 2–3 consecutive hues is good to go.</em></figcaption></figure><p>That’s it. The color scale is all-important when it comes to heatmaps, so remember to take good care of it. The utmost goal of a heatmap, or any other kind of visualizations, is to tell stories from the data. Just keep it clear and simple.</p><p>You can get <a href="https://vinci.bioturing.com/">BioVinci</a>, a drag-and-drop package for modern data visualizations, to quickly change a heatmap color scale. <a href="https://blog.bioturing.com/2018/05/08/how-to-build-a-hierarchical-clustering-heatmap-with-biovinci/">Here</a>’s a simple tutorial.</p><p>Have fun fine-tuning your heatmap!</p><p><em>Originally published at </em><a href="https://blog.bioturing.com/index.php/heatmap-color-scale/"><em>blog.bioturing.com</em></a><em> on September 24, 2018.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=75929663988b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to read PCA biplots and scree plots]]></title>
            <link>https://bioturing.medium.com/how-to-read-pca-biplots-and-scree-plots-186246aae063?source=rss-64e2fc4336f3------2</link>
            <guid isPermaLink="false">https://medium.com/p/186246aae063</guid>
            <category><![CDATA[pca-explained-simply]]></category>
            <category><![CDATA[pca-biplot]]></category>
            <category><![CDATA[scree-plot]]></category>
            <category><![CDATA[apc]]></category>
            <category><![CDATA[rna-seq]]></category>
            <dc:creator><![CDATA[BioTuring Team]]></dc:creator>
            <pubDate>Mon, 18 Jun 2018 23:39:39 GMT</pubDate>
            <atom:updated>2018-09-18T06:30:42.776Z</atom:updated>
            <content:encoded><![CDATA[<p>Principal component analysis (<a href="https://blog.bioturing.com/2018/05/12/tracing-origin-of-atacama-alien-skeleton-a-pca-step-by-step-example/">PCA</a>) has been gaining popularity as a tool to bring out strong patterns from complex biological datasets. We have answered the question “What is a PCA?” in <a href="https://blog.bioturing.com/2018/06/14/principal-component-analysis-explained-simply/">this jargon-free blog post</a> — check it out for a simple explanation of how PCA works. In a nutshell, PCA capture the essence of the data in a few principal components, which convey the most variation in the dataset.</p><h3>1. A PCA plot shows clusters of samples based on their similarity.</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/692/0*JOE-WM-0xvqV3EbG" /></figure><p><strong>Figure 1. PCA plot</strong>. For how to read it, <a href="https://blog.bioturing.com/2018/06/14/principal-component-analysis-explained-simply/">see this blog post</a></p><p>PCA does not discard any samples or characteristics (variables). Instead, it reduces the overwhelming number of dimensions by constructing principal components (PCs). PCs describe variation and account for the varied influences of the original characteristics. Such influences, or loadings, can be traced back from the PCA plot to find out what produces the differences among clusters.</p><h3>2. A loading plot shows how strongly each characteristic influences a principal component.</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/897/0*ATQugHVaM_w_WVSY" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/625/0*D_RWvtvarXuGAc2v" /></figure><p><strong>Figure 2. Loading plot</strong></p><p>See how these vectors are pinned at the origin of PCs (PC1 = 0 and PC2 = 0)? Their project values on each PC show how much weight they have on that PC. In this example, NPC2 and CHIT1 strongly influence PC1, while GBA and LCAT have more say in PC2.</p><p>Another nice thing about loading plots: the angles between the vectors tell us how characteristics correlate with one another. Let’s look at Figure 2.</p><ul><li>When two vectors are close, forming a small angle, the two variables they represent are positively correlated. Example: APOD and PSAP</li><li>If they meet each other at 90°, they are not likely to be correlated. Example: NPC2 and GBA.</li><li>When they diverge and form a large angle (close to 180°), they are negative correlated. Example: NPC2 and MAG.</li></ul><p>Now that you know all that, reading a PCA biplot is a piece of cake.</p><h3>3. PCA biplot = PCA score plot + loading plot</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/901/0*2CodrR55NBHKVOBW" /></figure><p><strong>Figure 3. PCA biplot</strong></p><p>You probably notice that a PCA biplot simply merge an usual PCA plot with a plot of loadings. The arrangement is like this:</p><ul><li>Bottom axis: PC1 score.</li><li>Left axis: PC2 score.</li><li>Top axis: loadings on PC1.</li><li>Right axis: loadings on PC2.</li></ul><p>In other words, the left and bottom axes are of the PCA plot — use them to read PCA scores of the samples (dots). The top and right axes belong to the loading plot — use them to read how strongly each characteristic (vector) influence the principal components.</p><h3>4. A scree plot displays how much variation each principal component captures from the data</h3><p>A scree plot, on the other hand, is a diagnostic tool to check whether PCA works well on your data or not. Principal components are created in order of the amount of variation they cover: PC1 captures the most variation, PC2 — the second most, and so on. Each of them contributes some information of the data, and in a PCA, there are as many principal components as there are characteristics. Leaving out PCs and we lose information.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/897/0*w-lrtcduVN80Mi2D" /></figure><p><strong>Figure 4. PCA scree plot</strong></p><p>The good news is, if the first two or three PCs have capture most of the information, then we can ignore the rest without losing anything important. A scree plot shows how much variation each PC captures from the data. The y axis is eigenvalues, which essentially stand for the amount of variation. Use a scree plot to select the principal components to keep. An ideal curve should be steep, then bends at an “elbow” — this is your cutting-off point — and after that flattens out. In Figure 4, just PC 1,2, and 3 are enough to describe the data.</p><p>To deal with a not-so-ideal scree plot curve, there are a couple ways:</p><ol><li>Kaiser rule: pick PCs with eigenvalues of at least 1.</li><li>Proportion of variance plot: the selected PCs should be able to describe at least 80% of the variance.</li></ol><p>If you end up with too many principal components (more than 3), PCA might not be the best way to visualize your data. Instead, consider other dimension reduction techniques, such as t-SNE and MDS.</p><p><strong>In summary:</strong> A PCA biplot shows both PC scores of samples (dots) and loadings of variables (vectors). The further away these vectors are from a PC origin, the more influence they have on that PC. Loading plots also hint at how variables correlate with one another: a small angle implies positive correlation, a large one suggests negative correlation, and a 90° angle indicates no correlation between two characteristics. A scree plot displays how much variation each principal component captures from the data. If the first two or three PCs are sufficient to describe the essence of the data, the scree plot is a steep curve that bends quickly and flattens out.</p><p>Looking for a way to create PCA biplots and scree plots easily? Try <a href="https://vinci.bioturing.com/">BioVinci</a>, a drag and drop software that can run PCA and plot everything like nobody’s business in just a few clicks.</p><p><em>Originally published at </em><a href="https://blog.bioturing.com/2018/06/18/382/"><em>blog.bioturing.com</em></a><em> on June 18, 2018.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=186246aae063" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Principal component analysis explained simply]]></title>
            <link>https://bioturing.medium.com/principal-component-analysis-explained-simply-894e8f6f4bfb?source=rss-64e2fc4336f3------2</link>
            <guid isPermaLink="false">https://medium.com/p/894e8f6f4bfb</guid>
            <category><![CDATA[pca-explained-simply]]></category>
            <category><![CDATA[apc]]></category>
            <category><![CDATA[biovinci]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[biology]]></category>
            <dc:creator><![CDATA[BioTuring Team]]></dc:creator>
            <pubDate>Thu, 14 Jun 2018 21:13:53 GMT</pubDate>
            <atom:updated>2018-08-01T03:16:10.156Z</atom:updated>
            <content:encoded><![CDATA[<p>As we are entering the era of Big Data, everyone and their moms seem to be talking about <a href="https://blog.bioturing.com/2018/05/12/tracing-origin-of-atacama-alien-skeleton-a-pca-step-by-step-example/">PCA</a>. All the papers you read mention PCA (with lots of jargon, of course). Half of the seminars you’ve been to this month touch on PCA. Your boss/collaborators suggest <a href="https://www.youtube.com/watch?v=6truZBwKEm8">trying PCA</a> on your data.</p><p>“What is this PCA thing that keeps running into me?” You ask.</p><p>Short for principal component analysis, PCA is a way to <em>bring out strong patterns from large and complex datasets</em>. For example, if you measure the expression of 15 genes from 60 mice, and the data come back as a 15×60 table, how do you make sense of all that? How do you even know which mice are similar to one another, and which ones are different? How do you know which genes are responsible for such similarities or differences?</p><p>PCA might be what you need to untangle that data mess. What it does is to take the expression data of 15 genes from each mouse and smoosh them down to one single dot that represents the expression profile of that mouse. One dot for one mouse. Sixty mice, sixty dots. The result will look like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/692/0*uo_XDQYV_8ewVSp_" /></figure><p>This plot makes it much easier to compare mice. Those with similar expression profiles will cluster together. In this plot, there are 3 clusters of mice. From here, you can trace back to find out which genes make the clusters different from one another (more on that later).</p><p>Neat! How do we go from that 15×60 table to this pretty plot?</p><h3>1. Principal components capture the most variation in a dataset</h3><p>Let’s not forget that what we are trying to do here, from the moment that the 15×60 table arrives and we stare at it in bewilderment, is to <em>compare the mice</em>. When there are too much information about them, it’s hard to do that. Let’s start with 2 genes to make things easier. Using those 2 genes as axes, we can plot their expression in 60 mice on a 2D plot, like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/692/0*vLVaAcb-jlgXylU8" /></figure><p>Here, each dot carries read counts of 2 genes from one mouse, and together they form a flat “cloud.” Principal component 1 (PC1) is a line that goes through the center of that cloud and describes it best. It is a line that, if you project the original dots on it, two things happen:</p><ol><li>The total distance among the projected points is maximum. This means they can be distinguished from one another as clearly as possible. We want to compare stuff, remember? If a line blurs data points together, it won’t help.</li><li>The total distance from the original points to their corresponding projected points is minimum. This means we have a representation that is as close to the original data as possible.</li></ol><p>In other words, the best line — our PC1 — must convey the <em>maximum variation</em> among data points and contain <em>minimum error</em>. These two things are actually achieved at the same time, because <a href="https://www.youtube.com/watch?v=FgakZw6K1QQ">Pythagoras said so.</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/692/0*fswM4uug48zvzRDc" /></figure><h3>2. PCA deals with the curse of dimensionality by capturing the essence of data into a few principal components.</h3><p>But we have 15 genes, not just 2. The more genes you’ve got, the more axes (dimensions) there are when you plot their expression. When it gets to about the fourth gene, you probably don’t want to plot this anymore, let alone 15 genes. 15 genes equal 15 axes! The cloud of dots is no longer flat, it is 15-D now. There is no way to look at it and make any sense out of it.</p><p>This is when the magic of PCA comes in. To create PC1, a line is anchored at the center of the 15-D cloud of dots and rotate in 15 directions, all the while acting as a “mirror,” on which the original 60 dots are projected. This rotation continues until the total distance among projected points is maximum. The rotating line now describes the most variation among 60 mice, and is fit to be PC1.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/625/0*XqrkXQoEQEIRe0Nm" /></figure><p><em>Source: stats.stackexchange.com</em></p><p>PC2 is the second line that meets PC1, perpendicularly, at the center of the cloud, and describes the second most variation in the data. (<a href="https://www.youtube.com/watch?v=mhNfhxJ7bFo">Computers can handle all this rotation and setting up PCs in a breeze</a>, lucky us!)</p><p>If PCA is suitable for your data, just the first 2 or 3 principal components should convey most of the information of the data already (more on this later). This is nice in several ways:</p><ol><li>Principal components help reduce the number of dimensions down to 2 or 3, making it possible to see strong patterns.</li><li>Yet you didn’t have to throw away any genes in doing so. Principal components take all dimensions and data points into account.</li><li>Since PC1 and PC2 are perpendicular to each other, we can rotate them and make them straight. These are the axes of our pretty PCA plot.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/692/0*P0rc9_1KWkThSAQB" /></figure><h3>3. Dimensions vary in the weights they have on each principal component.</h3><p>Now that we have the axes, how about dots? How do we smoosh 15 genes together to create a single dot for each mouse?</p><p>The original dots (mouse gene expression) are pinned in place by their values on all axes (genes). These dots are projected on PC lines, which are trying to stay as close as possible to all the dots — that’s how they achieve the best fit. For the PC lines, this is like trying to stand still while being pulled in 15 directions at once. Some pulling forces are stronger than others; in other words, dimensions have <em>different weights</em> in defining each PC.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/692/0*az1VOxSvwpTgAZvp" /></figure><p>In our example, all 15 genes participate in placing mice on PCs, but they weigh in differently. After<a href="http://www.sthda.com/english/articles/31-principal-component-methods-in-r-practical-guide/112-pca-principal-component-analysis-essentials/"> their weights are determined</a>, finding out where they put Mouse #1 on PC1 is simple math:</p><p>(gene1 read count * gene1 weight on PC1)</p><p>+ (gene2 read count * gene2 weight on PC1)</p><p>+ …</p><p>+ (gene15 read count * gene15 weight on PC1)</p><p>= PC1 value of Mouse #1</p><p>Mouse #1 has a spot on PC2, too. To calculate it, use the same formula with weights of genes on PC2 instead of PC1.</p><p>With a value of PC1 and a value of PC2, Mouse #1 now can be graphed as a dot on the PCA plot.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/300/0*TUaNbVAhOG8XJDPT" /></figure><p>This simple math can be written in a fancy formula that’s more appropriate for papers and seminars:</p><p>Let,</p><ul><li>Mouse #i :</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/300/0*beE5V7zyNAdhNqnj" /></figure><ul><li>Eigenvector #j:</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/300/0*zbIIf5KhlCV9HICQ" /></figure><ul><li>Principal component j-th of sample i:</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/100/0*9LFufeDT-14YOKzH" /></figure><p>Do this math again on all the mice, and they will each become a dot on the PCA plot.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/692/0*_riUr9E3l1Qgr7Jj" /></figure><p>Pretty! How do we read this?</p><h3>4. How to read a PCA plot</h3><ul><li>Mice that have similar expression profiles are now clustered together. Just glancing at this plot, we can see that there are 3 clusters of mice.</li><li>If 2 clusters of mice are different based on PC1, like the blue and orange clusters in this plot, such differences are likely to be due to the genes that have heavy influences on PC1.</li><li>If 2 clusters are different based on PC2, like the red and blue clusters, then the genes that heavily influence PC2 are likely to be responsible.</li><li>Keep in mind that PCs are ranked by how much they describe the data. PC1 reveals the most variation, while PC2 reveals the second most variation. Therefore, differences among clusters along PC1 axis are actually<em> larger</em> than the similar-looking distances along PC2 axis.</li><li>Is this plot meaningful? Check the<em> proportion of variance</em>, or the diagnostic scree plot. PCA is worthy if the top 2 or 3 PCs cover most of the variation in your data. Otherwise, you should consider other dimension reduction techniques, such as t-SNE and MDS.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*gAcOw3yZdODhvIUm" /><figcaption><em>Proportion of variance graphs, good and bad</em></figcaption></figure><p>To sum up, principal component analysis (PCA) is a way to bring out strong patterns from large and complex datasets. The essence of the data is captured in a few principal components, which themselves convey the most variation in the dataset. PCA reduces the number of dimensions without selecting or discarding them. Instead, it constructs principal components that focus on variation and account for the varied influences of dimensions. Such influences can be traced back from the PCA plot to find out what produces the differences among clusters.</p><p>To run a PCA effortlessly, try <a href="https://vinci.bioturing.com/">BioVinci</a>. This drag-and-drop software can perform any statistical analyses in just a few clicks. Check it out here: <a href="https://vinci.bioturing.com/">https://vinci.bioturing.com</a></p><p><em>Originally published at </em><a href="https://blog.bioturing.com/2018/06/14/principal-component-analysis-explained-simply/"><em>blog.bioturing.com</em></a><em> on June 14, 2018.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=894e8f6f4bfb" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>