<?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 Decode Box on Medium]]></title>
        <description><![CDATA[Stories by Decode Box on Medium]]></description>
        <link>https://medium.com/@decodebox?source=rss-2553b3805553------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*p09_GY8XsV4bxXOikuGuZQ.png</url>
            <title>Stories by Decode Box on Medium</title>
            <link>https://medium.com/@decodebox?source=rss-2553b3805553------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 16:40:14 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@decodebox/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[Normalizing Behaviors and RNASeq: A Tale of Good and Bad Practices]]></title>
            <link>https://medium.com/@decodebox/normalizing-behaviors-and-rnaseq-a-tale-of-good-and-bad-practices-72cb2a122cc9?source=rss-2553b3805553------2</link>
            <guid isPermaLink="false">https://medium.com/p/72cb2a122cc9</guid>
            <category><![CDATA[bioinformatics]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[bioinformatics-tools]]></category>
            <category><![CDATA[rna-sequencing]]></category>
            <category><![CDATA[data-experts]]></category>
            <dc:creator><![CDATA[Decode Box]]></dc:creator>
            <pubDate>Tue, 20 Aug 2024 18:30:28 GMT</pubDate>
            <atom:updated>2024-08-20T18:30:28.009Z</atom:updated>
            <content:encoded><![CDATA[<p>In bioinformatics, as with anything, using the wrong methods or tools, even with great resources, can lead to poor results. My somewhat unstructured path to learning kept me extra cautious with my daily work, constantly evaluating the results from a statistical lens. To an extent, there’s a certain joy in uncovering nuances that even seasoned scientists might overlook or misapply while dealing with statistical methods and data.</p><p>Previously, I have written blogs on how differential expression analysis for both bulk and single-cell RNASeq should be approached to yield statistically reliable outcomes. About a year ago, I noticed a similar gap in understanding of RNASeq data normalization.</p><p>Normalization is often treated as a routine step, yet many fail to fully appreciate the underlying statistical principles. <strong>The choice of normalization method can significantly impact downstream analysis, and selecting the right one requires a clear understanding of these fundamentals</strong>. Since I use my blog to both structure my understanding and share best practices with the community, I decided to address this often underestimated, yet crucial, topic with this blog.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/888/0*NY6tO11gOtz5Os2A" /><figcaption>Joker is just trying to make you aware</figcaption></figure><p>This blog serves as a more personalized guide to data normalization for RNASeq data. I hope that it will help uncover common issues and bring clarity to the subject.</p><h3>Why do we need Normalization ?</h3><p>Data normalization is a crucial step in any data analysis, particularly in bioinformatics. <strong>All data, whether biological or not, inherently contains variations — not only from actual differences but also from technical differences in measurement</strong>. These technical differences can make comparisons unreliable, both between features within a sample and across different samples. <strong>Normalization minimizes these technical variations while preserving the true biological differences</strong>. This ensures that features within a sample are comparable, and that samples as a whole can be reliably compared with each other.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Zy71dkk9LwV9qgg1" /><figcaption>Dummy data showcasing before normalization and after normalized version of the results</figcaption></figure><p>To illustrate what normalization can accomplish, consider the above plot. If we look at Gene A and Gene B as relevant genes — potentially up-regulated or down-regulated in tumors — based on the un-normalized data, Gene B might appear to be an up-regulated target. However, after normalization, it no longer seems up-regulated.</p><p>The reason for the apparent up-regulation in the un-normalized data could be due to <strong>technical factors such as differences in library size</strong>. In simpler terms, one sample might have captured more RNA than another, making it seem like there is a differential expression when, in reality, it’s just a technical artifact. Normalization corrects for this, ensuring that the observed differences are biologically meaningful.</p><p>Thus, normalization is a crucial step to ensure reliable results. <strong>If incorrect normalization methods are used in a specific context or analysis, they can lead to erroneous conclusions</strong>. The main section of this blog will discuss some of the common incorrect practices, but before diving into that, let’s take a quick look at the popular methods used for normalizing RNASeq data.</p><h3><strong>Types of Normalization Methods: Within Sample &amp; Across Sample</strong></h3><p>Broadly, the normalization methods for RNASeq can be categorized into two categories:</p><h4><strong>Within Sample Normalization</strong></h4><p>As the name suggests, within-sample normalization is aimed at making the gene expression within a sample comparable. For example, longer genes might have more mapped reads than shorter ones, making comparisons between them challenging.</p><p>Within-sample normalization methods typically <strong>correct for gene length and sequencing depth</strong> to account for these differences.</p><p><strong>Methods like CPM, FPKM, TPM, and others fall into this category. </strong>Within sample normalization methods should ideally be used to compare a gene with the others from a sample itself.</p><h4><strong>Across Sample Normalization</strong></h4><p>Across-sample normalization methods are more suited for comparing gene expression across different samples.</p><p>These methods are <strong>usually correct for sequencing depth and RNA composition</strong>, allowing for a more accurate comparison between samples.</p><p><strong>Methods like DESeq2 virtual sample normalization, TMM, and others are part of this category.</strong></p><h3>Why and how to Choose Normalization carefully.</h3><p><a href="https://doi.org/10.1186/1471-2105-11-94">Bullard, James H., et al</a> have shown that the<strong> impact of normalization methods on the DE results is higher than the impact of statistical tests used for calculating DE</strong>. This is a major highlight as usually, scientists try to optimize for DE statistical tests and not the normalization method over the data.</p><h4>A common bad practice still used routinely</h4><p>To provide more background on data normalization, the use of Within Sample Normalization became normalized (Pun Intended) among scientists aiming to create an absolute measure of RNA levels within sequenced samples. For instance, TPM (Transcripts Per Million) is often used as an absolute measure of how many transcripts exist in a sample per million total transcripts. <strong>While the concept sounds promising in theory, it has significant flaws.</strong></p><p>These methods fail to account for differences across sequencing protocols and samples. As <a href="http://www.rnajournal.org/cgi/doi/10.1261/rna.074922.120">Zhao et al. (RNA 2020; 26: 903–909)</a> demonstrated, <strong>variations in sequencing protocols — such as how the library is prepared for the same sample — can significantly impact measurements of different RNA classes normalized using TPM (results shared next from their publication)</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*JKDBAUWDgCZQJTMa" /><figcaption>Figure 1 from Shanron Zhao et al. RNA 2020;26:903–909</figcaption></figure><p>Furthermore, the assumption that TPM-normalized data is suitable for cross-sample comparisons is incorrect. TPM normalization only adjusts for sequencing depth and gene length, but it doesn’t account for RNA composition differences across samples. FPKM, RPKM, and similar measures are even less suited for such comparisons, as they fail to account for additional confounding factors.</p><p>Unfortunately, methods like CPM, TPM, FPKM, RPM, and RPKM have become commonplace because they were adopted by scientists in an attempt to quantify absolute RNA levels. <strong>This problem has been exacerbated by large consortia and bodies, such as TCGA, NCI, and GDC, which frequently use these normalization measures as default when making data available</strong> (although raw counts are also available, many researchers tend to use the within sample-normalized data provided by these consortia). <strong>Quoting directly from Zhao et al, RNA 2020 — “RPKM and TPM represent the relative abundance of transcripts in a sample but do not normalize for global shifts in total RNA contents.”</strong></p><h4>What do the experts and evidence have to say — Literature review</h4><p><a href="https://doi.org/10.1186/s12967-021-02936-w">Zhao et al, J. Transl Med 2021</a> evaluated the impact of different normalization methods on bulk RNASeq data. They examined both within-sample normalization methods (RPKM, FPKM, TPM, and z-score normalized TPM) and across-sample methods (DESeq2’s virtual reference sample-based normalization and EdgeR/NOISeq’s TMM) using metrics such as the coefficient of variation (CV) of gene expression, intraclass correlation, and sample clustering.</p><p>For their study, the authors used early-passage PDX data from the NCI PDMR (<a href="https://pdmr.cancer.gov/">https://pdmr.cancer.gov/</a>), which is known for its genomic and transcriptomic similarity to the original tumor. The dataset included 20 distinct PDX models with replicates, totaling 61 samples.</p><p>To assess the impact of normalization on gene expression variation, they calculated the median coefficient of variation (CV) across replicate samples for each PDX model. DESeq2 normalized counts and TMM-normalized data exhibited the lowest median CVs, with values ranging from 0.05 to 0.15, and performed comparably to each other. In contrast, within-sample normalization methods exhibited higher CVs compared to across-sample methods. TPM performed the worst, with median CVs ranging from 0.08 to 0.52, while FPKM performed better than TPM but still worse than across-sample normalized counts.</p><p>The replicates of the PDX model 475296–252-R showed significant variation in hierarchical clustering when using un-normalized data, making it an ideal candidate for further evaluation. The authors tested whether a z-score transformation or an additional TMM normalization step applied to TPM data could resolve this issue. Z-score transformed TPM data still resulted in separated replicates in hierarchical clustering, whereas TMM normalization clustered the replicates together. However, the TMM-normalized data showed a coordinated shift for highly expressed genes. Additionally, the median CV for all genes in TMM-normalized TPM data was higher than that of TMM-normalized count data.</p><p><a href="https://doi.org/10.1093/bib/bbs046">Dillies, Marie-Agnès, et al</a> have done a comparison of five across sample methods — Total Counts, Median Counts, Quantile, Upper Quartile, DESeq’s Median of ratios based method, and Trimmed Mean of M-values (TMM) from EdgeR, and one within-sample normalization method — RPKM. They did the evaluation across real and simulated datasets, focusing on robustness and stability in various conditions.</p><p>For real data, two criteria were used:</p><ol><li>Qualitative characteristics of normalized data (e.g., distribution of counts and intra-group variability)</li><li>Results from differential expression (DE) analyses.</li></ol><p>All methods perform similarly in datasets with low variability, but significant differences emerged when library sizes varied. Methods like DESeq and TMM consistently showed better stabilization of read count distributions and minimized intra-group variance, particularly in complex datasets. RPKM and TC performed poorly in stabilizing variability and should be avoided for DE analysis.</p><p>For simulated data, DESeq and TMM stood out for controlling false-positive rates and maintaining power, even in challenging conditions involving high-count genes. Other methods, like TC, UQ, Med, and Q, struggled with inflated false-positive rates in these settings.</p><p>Overall, <strong>DESeq and TMM demonstrated strong performance across various datasets</strong>. However, these methods rely on the assumption that most genes are not differentially expressed, which may not always hold true. Despite this, DESeq and TMM remain recommended due to their robustness and ease of implementation via Bioconductor packages.</p><p>This concludes that the <strong>Within-Sample normalization methods should be used much more cautiously and ideally should be discouraged for across-sample comparisons.</strong> Let’s talk more about the Across sample normalization methods with a study using only them and their suggestions based on the same.</p><p><a href="https://doi.org/10.1155/2015/621690">Zyprych-Walczak, J., et al BioMed Res 2015</a>, have done a comparison of multiple across sample methods — Trimmed Mean of M-values (TMM) (EdgeR), Upper Quartile, Median of ratios (DESeq), Quantile (EBSeq), and PoissonSeq (PoissonSeq 3) normalization. They used two publicly available datasets and one internal dataset for the evaluation. The comparison of the methods was done based on four evaluation criteria:</p><ol><li>Bias and variation of Housekeeping Genes post normalization with the methods: The lower the bias and variance of these genes, the higher is the method ranked.</li><li>The number of Common DEGs (Done using the EdgeR DE method) identified after normalization from the different methods.</li><li>Discriminant analysis based on the classification ability of DEGs for each dataset.</li><li>Sensitivity and Specificity of the classification exercise</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/866/1*QUHWsy2AL5DKXcOmvg8Udg.png" /><figcaption>Table 7 from <a href="https://doi.org/10.1155/2015/621690">Zyprych-Walczak, J., et al BioMed Res 2015</a>. It highlights the overall summary ranks of different methods as per different criteria over the AML dataset used in the study.</figcaption></figure><p>The study has many nuances and I recommend going through the study to have a more thorough understanding. For the sake of the length of the blog, I am summarizing their findings.</p><ol><li>Different methods had very different results and ranks for different measures of evaluation.</li><li>TMM performed poorly in most evaluation criteria. There seem to be conflicting results about TMM in the author’s study in comparison to studies done by <a href="https://doi.org/10.1093/bib/bbs046">Dillies, Marie-Agnès, et al</a> and <a href="https://doi.org/10.4161/cib.25849">Maza, Elie, et al.</a> which the authors acknowledge could be due to differences in evaluation criteria.</li><li><strong>DESeq Median-based method performed well in most cases and was ranked higher in most evaluation criteria</strong>.</li></ol><p>The authors presented a great way to evaluate which methods and results to choose finally for analysis, I am listing down a summarized version of their workflow for the evaluation.</p><p>To determine the optimal normalization method for a specific dataset, <strong>they and we also recommend the following workflow</strong>:</p><ol><li><strong>Normalize the Data:</strong> Apply various normalization methods to the dataset.</li><li><strong>Evaluate Bias and Variance:</strong> Calculate and rank the methods based on their bias and variance.</li><li><strong>Perform Differential Analysis:</strong> After normalization, conduct differential expression analysis and compare the differentially expressed gene (DEG) lists produced by each method.</li><li><strong>Use Control Genes:</strong> Select positive and negative control genes to evaluate the sensitivity and specificity of each method, and rank them accordingly.</li><li><strong>Assess Prediction Errors:</strong> Calculate the mean prediction errors using classifiers for DEGs found by each method and rank them.</li><li><strong>Visualize Overlap:</strong> Create Venn diagrams or balloon plots to visualize the overlap of DEGs and rank methods based on the number of common genes.</li><li><strong>Choose the Best Method:</strong> Summarize the rankings to select the most suitable normalization method for the dataset.</li></ol><h3>Conclusion</h3><p>Normalization is a critical step in RNASeq data analysis that directly influences the reliability of downstream results, including differential expression analyses. The choice of normalization method should be made carefully, considering both the biological context and technical variations within and across samples. Incorrect or inappropriate normalization can lead to biased conclusions and misinterpretation of biological phenomena.</p><p>To end the blog, I have a set of summary guidelines that can help you get reliable results with your analysis based on proper normalization of the data.</p><h4><strong>Suggestions/Guidelines:</strong></h4><ol><li><strong>Understand Your Data:</strong> Evaluate your dataset’s characteristics (e.g., library size, RNA composition) before choosing a normalization method. This ensures the method aligns with the specific nuances of your data.</li><li><strong>Choose Normalization Based on Purpose:</strong> Use within-sample normalization (e.g., TPM, FPKM) only for comparing expression levels within a sample. For cross-sample comparisons, rely on across-sample methods like DESeq2 or TMM.</li><li><strong>Evaluate Performance:</strong> After normalization, assess the bias, variance, and impact on differential expression results. Using control genes and performing comparative analyses will help verify the effectiveness of the chosen method.</li><li><strong>Test Multiple Methods:</strong> If uncertain, apply different normalization methods and compare their results using metrics like coefficient of variation and classification accuracy. Choose the method that minimizes technical noise while preserving biological signals.</li><li><strong>Be Cautious with Default Methods:</strong> Consortium-provided normalized data may not always be suitable for specific analyses. When possible, work with raw counts and apply your normalization strategies tailored to your study.</li></ol><p>These guidelines will help ensure that your RNASeq data analysis produces biologically meaningful and statistically reliable results.</p><h3>References:</h3><ol><li>Bullard, James H., et al. <a href="https://doi.org/10.1186/1471-2105-11-94">“Evaluation of statistical methods for normalization and differential expression in mRNA-Seq experiments.”</a> <em>BMC bioinformatics</em> 11 (2010): 1–13.</li><li>Zhao, Shanrong, Zhan Ye, and Robert Stanton. <a href="http://www.rnajournal.org/cgi/doi/10.1261/rna.074922.120">“Misuse of RPKM or TPM normalization when comparing across samples and sequencing protocols.”</a> <em>Rna</em> 26.8 (2020): 903–909.</li><li>Zhao, Yingdong, et al. “<a href="https://doi.org/10.1186/s12967-021-02936-w">TPM, FPKM, or normalized counts? A comparative study of quantification measures for the analysis of RNA-seq data from the NCI patient-derived models repository.</a>” <em>Journal of translational medicine</em> 19.1 (2021): 269</li><li>Dillies, Marie-Agnès, et al. “<a href="https://doi.org/10.1093/bib/bbs046">A comprehensive evaluation of normalization methods for Illumina high-throughput RNA sequencing data analysis.</a>” <em>Briefings in bioinformatics</em> 14.6 (2013): 671–683.</li><li>Zyprych-Walczak, J., et al. <a href="https://doi.org/10.1155/2015/621690">“The Impact of normalization methods on RNA‐seq data analysis.”</a> <em>BioMed research international</em> 2015.1 (2015): 621690.</li><li>Maza, Elie, et al. <a href="https://doi.org/10.4161/cib.25849">“Comparison of normalization methods for differential gene expression analysis in RNA-Seq experiments: a matter of relative size of studied transcriptomes”</a> <em>Communicative &amp; integrative biology</em> 6.6 (2013): e25849.</li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=72cb2a122cc9" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Pen to the Pyramid Peak]]></title>
            <link>https://medium.com/@decodebox/pen-to-the-pyramid-peak-7cef8b3e7b68?source=rss-2553b3805553------2</link>
            <guid isPermaLink="false">https://medium.com/p/7cef8b3e7b68</guid>
            <category><![CDATA[career-development]]></category>
            <category><![CDATA[learning]]></category>
            <category><![CDATA[career-advice]]></category>
            <category><![CDATA[science-writing]]></category>
            <category><![CDATA[nonfiction-writing]]></category>
            <dc:creator><![CDATA[Decode Box]]></dc:creator>
            <pubDate>Wed, 03 Apr 2024 12:23:27 GMT</pubDate>
            <atom:updated>2024-04-03T13:17:12.670Z</atom:updated>
            <content:encoded><![CDATA[<p>As I started a blog space, wrote a few chapters for an academic book on antimicrobial resistance (soon to come out), wrote multiple topical blogs for my organization, and wrote for a personal interview recently, I appreciate the importance of non-fiction writing as a professional skill even more.</p><blockquote>I am highly inspired by the line — <strong>Keeping thoughts to yourself can be a disservice to the world. If you have something important to say and you say it well, you send strangers down paths badly needed</strong> from <strong>Julian Shapiro’s</strong> <a href="https://www.julian.com/guide/write/intro">Writing better — Why write non-fiction?</a> handbook (I suggest you give it a go),</blockquote><p>In this process of writing and reading other non-fiction content such as the one from Julian Shapiro, as a professional especially in a career like research and bioinformatics, I found my reasons to why writing non-fiction matters.</p><p>It is my personal belief that everyone should pursue meaningful work and chase impact creation through it. While what defines a career fulfilling that, along with motivation and drive to pursue one, might be different for different folks, it is important to find your edge and leverage which helps you fulfill your chase. Investing in your skills continuously is an important edge in my opinion. Let me explain my reasoning for saying that.<em> </em><strong><em>If you consider the skill involvement in the growth of your career like a pyramid, a strong foundation is built on a strong suite of hard skills</em></strong>. As a biological/bioinformatics researcher, this might include an understanding of the biological systems, proficiency with computational methods and programming languages, etc. <strong><em>The peak, however, will be achieved using a set of soft skills over the hard skill foundation</em></strong>. Soft skills like articulation, creative and critical thinking, problem-solving attitude, ability to learn new things, attention to detail, etc. are all aspects of one’s personality (trainable) that impact how far one will reach in one’s career. Of course, this consideration is regarding one’s skills for one’s professional growth and success and there are other factors beyond skills that play a role.</p><p>As much as it is important to build a strong foundation and keep on ever-expanding that, it is also important to invest in one’s soft skills. If you may ask how can one build soft skills, I might not have a structured answer for you right now (but I will surely bring one later), however, if the question becomes, how can one better their soft skills, that’s where the subject topic — investing in non-fiction writing comes. Engaging in non-fiction writing can result in holistic growth across different aspects of one’s soft skill set. Below, I articulate some of my thoughts on this subject matter.</p><h3>Structured Thinking</h3><p>While supervisors and leaders in a lab or an organization may value the structured documents, articles, and presentations we produce, the true value lies in the structure it brings to our minds. Our brains are repositories of knowledge, ideas, and thoughts, often scattered and unorganized. However, through the act of writing, we impose a structure upon this chaos. For instance, when I wrote my first blog on <a href="https://medium.com/@decodebox/is-your-differential-expression-result-lying-to-you-379784ddee51">differential expression analysis for RNA-Seq data</a>, I had a grasp of the subject matter, but my understanding was fragmented. It wasn’t until I structured and articulated my thoughts in writing that I realized the transformative power of organized thinking. <strong>Writing a blog on it allowed me to create a mental decision tree, where each branch represented a piece of knowledge or an idea, leading to a final decision or conclusion.</strong> This structured approach not only enhanced my understanding of the topic but also empowered me to navigate complex problems with clarity and confidence. Yes, I added more pieces to the structure by reading more on the subject but the act of writing it cemented a structure to my thought process.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/1*spW0rdJx-yr7pqXFAyHXEA.jpeg" /><figcaption>Thanos writes and gets the structured thinking</figcaption></figure><h3><strong>Portray Expertise and Confidence</strong></h3><p>As a bioinformatics scientist, I tackle intriguing challenges from researchers from industry and academia worldwide. It has been my experience that portraying your confidence to your clients/collaborators helps you seize opportunities to solve even more complex problems. Sometimes with limited interaction with your collaborators, showcasing your expertise becomes an even more difficult challenge. Writing non-fiction becomes an edge for you in that case. Even with my limited writing, I have garnered a lot of confidence and attention. When discussing topics or addressing problems, having a blog post or written reference allows me to swiftly explain my points. <strong>Sharing my understanding of a deep bioinformatics topic through my blogs has often solidified my impression as an expert.</strong> I know of an instance when my organization was able to create a strong impression with a potential collaboration due to a specific <a href="https://www.elucidata.io/blog/navigating-the-single-cell-ship-through-the-pseudo-bulk-route">single-cell methodology blog</a> written by me.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/1*icPOvqmtL_vWQml1w3T98w.jpeg" /><figcaption>Not discounting the weeks I spent in studying to write the blog</figcaption></figure><h3>Deeper Learning</h3><p>The process of engaging in structuring information through writing is the best path to deepening learning. As a scientist, there are topics out there that I am interested in understanding more. I might read something or even audit a course, but due to the limited attention span at the moment, it won’t be an immersive learning experience. <strong>However, the act of writing deepens understanding and cohesiveness in structure</strong>. I also realize the gaps in my current knowledge and fill those up. Recently I wrote a blog on <a href="https://www.elucidata.io/blog/comparison-of-single-cell-data-visualization-tools-part-2">single-cell visualization applications</a>. I had a good idea about the applications out there but I didn’t have an idea of the offered capabilities and suitability for various users and research contexts. As I wrote the blog, I comprehensively reviewed 30 applications, and based on their capabilities, I arrived at 5 which covers different use cases for different researchers.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/583/1*veQJPbCnyV-M5bbfhrTf2A.jpeg" /><figcaption>Nope I didn’t, it is not always</figcaption></figure><h3>Pulling Opportunities</h3><p>Organizations, project owners, and leaders are seeking individuals out there who know how to solve problems. They search for people who have a personality to impact the space positively. In a landscape where one’s peers might have equivalently good hard skills, a generic resume, and LinkedIn profile don’t help (of course invest time in improving on those as your non-fiction content space). Writing non-fiction for your thoughts on topical subjects, hobbies, passions, interests, and ideas relays your experiences, personality, nature, and motivations very strongly. The topical content in your domain, helps a hiring manager understand your skills and your ways of demonstrating them. The other pieces help them understand your personality and what drives you. <strong>In a world where recruiters are pressed for time, putting yourself out there through your writing helps them make an easy choice in your favor.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/888/1*S5CANCL0AY78Qe1ihmmh1Q.jpeg" /><figcaption>Well only if you write non-fiction</figcaption></figure><h3>Know Yourself Better</h3><p>Understanding oneself deeply not only enriches personal life but also fuels professional growth. A person who can introspect and improve can seize the world. Writing is a great way to do that. As a simple anecdote, while writing this blog, I reached out to peers and ChatGPT for feedback. A recurring piece of advice I received is using better transitions while writing. Reflecting on this feedback, I realized a similar need for smoother transitions sometimes in my verbal communication. This constructive criticism underscores the significance of writing in refining my communication skills, a valuable insight I might have overlooked otherwise. This highlights how writing offers insights into our communication styles and areas for development. Writing unveils various facets of our personality. For instance, a person’s obsessive focus on certain aspects (let’s say on technical details) while writing at the expense of clarity while writing often mirrors similar tendencies in verbal communication. <strong>Engaging in writing fosters a deeper understanding of our thought processes, and our ability to convey ideas effectively, and participate in conversations impact fully.</strong> Through introspection and inquiry, writing becomes a tool for both understanding and uncovering aspects of ourselves. A part of understanding yourself through any medium has a spiritual and personal side that cannot be discounted. Writing serves as a reflective practice that not only enhances professional skills but also nurtures our inner selves.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/670/1*qmQAmUTTkN7k2W0oA9p4Gw.jpeg" /><figcaption>Yes I did</figcaption></figure><p>Writing something other than the bioinformatics topical content is a first in a long time for me but writing non-fiction overall has been a steady source of soft skill growth for me. I know it is going to contribute to my professional growth over the years to come. On a personal note, I am eager to share my insights and ideas, knowing they at least a few of them have the potential to inspire others. Expressing oneself through writing serves as a profound means of connecting with others and leaving a lasting impact. While writing fiction can be daunting and may not resonate with everyone, non-fiction writing offers a versatile avenue accessible to all, enabling individuals to showcase their expertise and carve out successful careers. While academic papers and conference presentations hold high significance as non-fiction writing for researchers, aspiring to find other avenues to share one’s wonderful thoughts and inspirations with the world is equally imperative. <strong>I know researchers and entrepreneurs in deep tech who have shared their inspiration to discover and build new principles, fields, and tech through their non-fiction writing and other pieces of content with the world and me.</strong></p><blockquote>In the end<em>, </em><strong><em>I will say show yourself to the world through your work and amazing things will start happening in your life.</em></strong></blockquote><p>P. S</p><p>I have not engaged in writing non-fiction other than the bioinformatics topical content for a long while. I want to use this listicle to give a shout out to <a href="https://www.linkedin.com/in/arvindhsundar/">Arvindh Sundar</a> as his sessions gave me a final push to share more of my ideas through non-fiction writing. He is a marketing coach working with organizations to improve their marketing game. I met him recently as a part of training sessions organized by my organization. I like the attitude with which he shares his personality and thoughts with the world. I learned about a framework to presenting well, got introduced to Julian Shapiro’s guide to writing better and got a big reading list. He has a tradition to ask for a ‘love letter’ in acknowledgement of learning new things and ideas through his coaching. Through this post script, I am abiding by his request. My biggest take away from his sessions is that everyone’s thoughts have an audience, it is criminal to not share them with the world and writing is a strong medium to do that 😊.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7cef8b3e7b68" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Single cell differential expression: sham or incorrect practices ?]]></title>
            <link>https://medium.com/@decodebox/single-cell-differential-expression-sham-or-incorrect-practices-bea469ae75d8?source=rss-2553b3805553------2</link>
            <guid isPermaLink="false">https://medium.com/p/bea469ae75d8</guid>
            <category><![CDATA[omics-data-analysis]]></category>
            <category><![CDATA[single-cell-sequencing]]></category>
            <category><![CDATA[bioinformatics]]></category>
            <category><![CDATA[differential-expression]]></category>
            <category><![CDATA[single-cell-analysis]]></category>
            <dc:creator><![CDATA[Decode Box]]></dc:creator>
            <pubDate>Sat, 04 Feb 2023 13:49:06 GMT</pubDate>
            <atom:updated>2023-02-04T14:34:05.082Z</atom:updated>
            <content:encoded><![CDATA[<h3>Single cell differential expression: sham or incorrect practices ?</h3><p>The next phrase that continues after this sentence is quite harsh but it is also the one that I have routinely heard about single cell expression data from different scientists over 3 years of working with the data. <strong>Single cell differential expression is a sham !!</strong> This might sound harsh for a method, instrumental even for the basic cell type identification step, however at times I also have felt similarly. My rational mind however prompted me to see this problem either as a data-method misfit or a lack of understanding of the proper use of methods.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Z9MN0VPBJDP89pp5" /><figcaption>Photo by <a href="https://unsplash.com/@pawel_czerwinski?utm_source=medium&amp;utm_medium=referral">Pawel Czerwinski</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Before talking about the single cell expression methods, I would first like to discuss the issues that underlie the above feeling:</p><ol><li><strong>No matter what, the p-values never seem to be non-significant</strong>: Even if there are minimal differences between the two cell populations, the p-value that one would get even after adjusting for FDR would be very low and usually never non-significant.</li><li><strong>High false-positive values in fold change:</strong> If one tries to compare the expression of cell types in two different conditions, often you get a lot of differentially expressed genes, however in often cases most genes don’t show a strong association with these states if compared based on expression or correlation.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*x8OZoZD0J6MBLI3X6QN2cw.png" /><figcaption>Top differentially expressed (calculated using MAST) genes (Partial list) by dividing CD8+ T cells into two groups based on the high and low expression of the gene of interest in NSCLC tumors. The single cell data was taken from GEO dataset GSE99254 submitted by Guo et al, Nature Medicine, 2018</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rFLf_wsoJmxAqDI5ZwjQOQ.png" /><figcaption>Correlation coefficient of the identified differentially expressed genes in the previous figure with the gene of the interest in CD8+ T cells in NSCLC tumor. The single cell data was taken from GEO dataset GSE99254 submitted by Guo et al, Nature Medicine, 2018</figcaption></figure><p>As we can see from the conflicting results in the above figures, genes that have been identified as differentially expressed (by splitting CD8+ T cells into two groups based on the high and low expression of a gene of interest) didn’t show a strong correlation (Pearson) with each other and gene of interest. This is one of the several example off-cases that I have observed while working with single cell DE.</p><p>Let’s look if others have also faced similar issues with single cell differential expression. Multiple studies (Wang et al, BMC Bioinformatics, 2019; Dal Molin, Alessandra, Giacomo Baruzzo, and Barbara Di Camillo, 2017, Frontiers in Genetics; Das et al, 2021, Genes; Squair et al, Nature Communications, 2021) have highlighted that the performance of single cell DE methods is subjective and data dependent. Additionally, coherence between the popular DE methods can be very variable as shown by Wang et al, BMC Bioinformatics, 2019.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xDaN6jqdDnvi20IgfnOBGg.png" /><figcaption>Numbers of pairwise common DE genes tested by top 1000 genes in real data as tested by Wang et al, BMC Bioinformatics, 2019 (Figure 5)</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Xv7A0kLdX6KR2waA-eZLNQ.png" /><figcaption>Numbers of pairwise common DE genes tested by adjusted <em>p</em>-value&lt; 0.05 in real data as tested by Wang et al, BMC Bioinformatics, 2019 (Figure 6)</figcaption></figure><p>From these results, it is evident that there is very low concordance between different methods.</p><p>As I started exploring methods, their comparative analysis, and correctness, I realized that there are essential inherent properties of single cell data that give rise to challenges in getting accurate differential expression results:</p><ol><li><strong>High sparsity: </strong>A easy to distinguish characteristic of single cell data is the high proportion of zeros in the expression. These zeroes can be considered as actual data points of no expression of a gene and/or false/missed expression due to technical causes like the sensitivity of the instrument, half-life of mRNA, and other reasons.</li><li><strong>Multimodality: </strong>With bulk sequencing data, it is comparatively easy to classify a gene as differentially expressed or not as the distribution of genes is usually uni-modal (distribution having one clear peak) in bulk sequencing data. However, due to cellular heterogeneity and different cell types, the single cell gene expression is mostly multi-modal.</li><li><strong>Data size:</strong> While our capability to get expression profiles of millions of cells has greatly increased our understanding of biological systems, the high number of subjects (cells in this case) comes with its own set of data science challenges.</li></ol><p>Squair et al, Nature Communications, 2021 tried to identify ways to confront high false positives in single cell differential expression. They reasoned that datasets, where the same population of purified cells has been sequenced for both bulk and single cell sequencing, can be used for understanding the discrepancies. They found 18 such datasets in the public domain and considered them as gold standard datasets for comparison. Collectively they identified that</p><p><strong>Pseudo bulk based method performs better:<br></strong>Pseudo-bulk methods perform aggregation of expression of cells from different groups within a biological replicate. By doing so the methods thereby reduce the overall zero inflation. Additionally, Murphy et al, Nature Communications, 2022 also showed that pseudo-bulk methods tend to perform better than other methods for single cell DE.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/664/1*v1Ggzewm5fp_utR7Qk61VA.jpeg" /><figcaption>Area under the concordance curve (AUCC) for fourteen DE methods in the eighteen ground-truth datasets as tested by Squair et al, Nature Communications, 2021 (Figure 1C)</figcaption></figure><p>However one can also reason that the performance of the pseudo bulk methods is only comparatively better than the others and still suffers from low performance as evident from the low AUCC for even the pseudo bulk based methods.</p><p><strong>DE methods are biased towards highly expressed genes:<br></strong>Comparing the non-aggregation with the pseudo bulk based methods for single cell DE, the authors identified that false positives identified by the former set of methods are usually amongst the high-expressing genes. Thus even if the actual difference between two groups of cells amongst highly expressed genes is minimal, they can still be falsely identified as differentially expressed genes. Conversely, false negatives overlooked by non-aggregation-based methods are usually lowly expressed genes. To test whether the aggregation has any role to play, the authors avoided the aggregation step in the pseudo bulk methods and considered each cell as a sample to compute differential expression. This gave rise to higher false positives which were from a set of highly expressed genes thus validating their hypothesis.</p><p><strong>DE analysis should account for biological differences between replicates:<br></strong>Single cell DE can allow us to compare groups of cells from different biological replicates and groups. However if the groups of cells are formed without considering the sample origin and replicate from which they are originating, even pseudo-bulk based methods perform worse. The authors mixed cells coming from different replicates into different groups and found that the accuracy of the pseudo-bulk based methods was lost.</p><p><strong>Ending Notes</strong></p><p>From all the studies that I read through, it is evident that we don’t have one method that can fit well for all datasets. Similarly, even the best of methods tend to have some tradeoffs for a set of genes. Based on these studies, the best practices that can be adopted currently for single cell DE are</p><ol><li>Use pseudo-bulk based methods for single cell DE as opposed to model-based or even other non-aggregation based methods.</li><li>Make a comparison between cells or cell types by considering the biological replicate from which they originate.</li><li>Validate the differentially expressed genes with the highest fold change based on their expression/correlation in different groups.</li><li>If trying to identify markers/signatures for individual groups of cells, validate the DE results with statistical or machine learning methods especially optimized to work with high sample size and zero-inflated data.</li></ol><p>There are more than 100 DE methods for single cell available right now but only a handful of them are used across most studies. The inherent properties of single cell data require us to look at the statistical base underlying the methods in a new light. This will help us repurpose existing methods/identify new methods to calculate accurate single cell differential expression.</p><p><strong>References</strong></p><ol><li>Wang, T., Li, B., Nelson, C.E. <em>et al.</em><a href="https://doi.org/10.1186/s12859-019-2599-6"> Comparative analysis of differential gene expression analysis tools for single-cell RNA sequencing data</a>. <em>BMC Bioinformatics</em> <strong>20</strong>, 40 (2019).</li><li>Squair, J.W., Gautier, M., Kathe, C. <em>et al.</em> <a href="https://doi.org/10.1038/s41467-021-25960-2">Confronting false discoveries in single-cell differential expression</a>. <em>Nat Commun</em> <strong>12</strong>, 5692 (2021).</li><li>Murphy, A.E., Skene, N.G. <a href="https://doi.org/10.1038/s41467-022-35519-4">A balanced measure shows superior performance of pseudobulk methods in single-cell RNA-sequencing analysis</a>. <em>Nat Commun</em> <strong>13</strong>, 7851 (2022).</li><li>Dal Molin A, Baruzzo G and Di Camillo B. <a href="https://doi.org/10.3389/fgene.2017.00062">Single-Cell RNA-Sequencing: Assessment of Differential Expression Analysis Methods.</a> <em>Front. Genet</em>. 8:62 (2017).</li><li>Das S, Rai A, Merchant ML, Cave MC, Rai SN. <a href="https://doi.org/10.3390/genes12121947">A Comprehensive Survey of Statistical Approaches for Differential Expression Analysis in Single-Cell RNA Sequencing Studies</a>. <em>Genes</em>. 12(12):1947 (2021).</li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bea469ae75d8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Is your differential expression result lying to you ?]]></title>
            <link>https://medium.com/@decodebox/is-your-differential-expression-result-lying-to-you-379784ddee51?source=rss-2553b3805553------2</link>
            <guid isPermaLink="false">https://medium.com/p/379784ddee51</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[bioinformatics]]></category>
            <category><![CDATA[differential-expression]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[computational-biology]]></category>
            <dc:creator><![CDATA[Decode Box]]></dc:creator>
            <pubDate>Sat, 03 Dec 2022 12:46:03 GMT</pubDate>
            <atom:updated>2022-12-03T14:57:01.245Z</atom:updated>
            <content:encoded><![CDATA[<h3>Is your differential expression result lying to you ?</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*iKZhubD7xH_vCVFV" /><figcaption>Photo by <a href="https://unsplash.com/@cdd20?utm_source=medium&amp;utm_medium=referral">愚木混株 cdd20</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>About 6 months ago, I was working late at the night with a lot of differential expression comparisons over multiple datasets for a project. I realized that something was fishy with my results. Most of the hits across different comparisons didn’t seem to have any actual expression difference between the cohorts. At first, I thought that I made a mistake while scaling or during other processing steps of the RNA sequencing datasets but I traced back and found that there wasn’t any issue with processing. This irked me so much that I decided to pull an all-nighter and found much-required clarity regarding differential expression methods and data.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2uS_pt5_arN5M3O3uMhP4w.png" /><figcaption>Dummy data showcasing the inflated false positives based on average difference in normalized expression and logFC</figcaption></figure><p>I realized on that day that when it comes to solving a biological question using computational methods a sizeable portion of the community is mostly focused on the methods themselves. It’s usually perceived that a standard method when applied can solve all the cases. Reality is much farther from it as the method plus context together decide the closeness of the outcome to the true picture.</p><p>In the case of differential expression, the general perception is that relatively new methods (DESeq2, edgeR, and others) would work well and would identify a gene to be differentially expressed (DE) with very high accuracy. However, accurate prediction of DE genes is equally if not more dependent on data as the method.</p><h3>Background</h3><p>Differential expression methods were broadly created to study differences between two different biological states (Tumour vs Normal, Cirrhotic liver vs Non-Cirrhotic liver, Covid-19 Severe vs Asymptomatic cases) based on the expression of genes. Broadly with microarrays, non-parametric methods such as t-test and Wilcoxon Rank Sum test were used and later specific methods such as limma evolved. They didn’t consider data to be distributed in a certain way and relied on general hypothesis testing to identify genes of interest based on effect size (fold change) and statistical scores (p-values). Later on, with RNA-Seq, more parametric methods came into the picture such as DESeq, DESeq2, edgeR, and others. The primary change in the way these methods work is based on understanding that expression data in general has a Negative Binomial (NB) or Poisson distribution. This is based on the fact that most genes in two biologically different states have similar expression and there are very few genes that have a difference in those two states.</p><p>With this assumption, the belief came to be that methods such as Deseq2 and edgeR are advanced and work perfectly almost 100% of the time. However, if that’s the case, why did I get a discrepancy in the actual expression difference and fold change? I searched for literature to understand differential expression in depth.</p><h3>Reasons</h3><p>Data distribution to the good extent is the factor affecting the presence of inflated false positives in your DE results. The primary assumption that the gene expression has a NB or Poisson distribution isn’t correct in all cases. There are different backgrounds over which an expression distribution can loose its NB/Poisson nature.</p><h4><strong>Large sample size:</strong></h4><p>Li et al, Genome Biology 2022 explored the reliability of popular DE methods Deseq2 and edgeR and found huge discrepancies in the results between the two methods and also observed high false discovery rates when working with population data of large sizes. They found that genes falsely identified (with negative control data having no true difference between two conditions) as DE genes with large populations didn’t fit well into the NB distribution. Both methods had exaggerated false positives when working with large-sized population data as the genes might not follow an NB distribution in these cases. On the other hand, the Wilcoxon rank-sum test had much better FDR control as it assumes a different null hypothesis that a gene has equal chances of being differentially negatively or positively expressed between two conditions.</p><p>The paper also observed that FDR control for a method depends a lot on the sample size. If the sample size is less than 8, then parametric methods might be more reliable in controlling the FDR for the DE data, however when the sample size is more than 8, then non-parametric methods have much more power (probability of rejecting the null hypothesis) and FDR control.</p><p>The author has also written a very succinct article describing their findings on <a href="https://towardsdatascience.com/deseq2-and-edger-should-no-longer-be-the-default-choice-for-large-sample-differential-gene-8fdf008deae9">Towards Data Science</a>. The article helped me understand the reliability of Differential Expression methods in specific cases.</p><h4>True difference:</h4><p>Another study by Rocke et al, bioRxiv, 2015 (Pre-print and not peer-reviewed) explored the possible reasons for the cause of excess FDR in the DE results for different methods when there isn’t a true difference between the compared groups with simulated and public data. They also found out that in this case when there isn’t any true difference between the compared groups, the distribution doesn’t follow negative binomial fit and hence parametric methods (DESeq, DESeq2, edgeR) fail miserably in terms of FDR control and power.</p><h3>Avoiding False Positives</h3><p>Thus to trust that your DE results are telling you the truth, one need to play the data game. A few important considerations can help you avoid false positives in your data.</p><ol><li>Always plot the distribution of the data before doing DE: If the distribution shows a NB/Poisson nature, use parametric methods such as DESeq2, edgeR and others to calculate DE results, if not, follow non-parametric methods.</li><li>Check sample size: If the sample size is greater than 8, always use a non-parametric method.</li><li>Use Principle Component Analysis (PCA): Make a PCA plot and check the difference in localization of samples of two different cohorts. If you are not getting a good difference then there isn’t a strong true difference in their expression profiles. If you have a lot of DE results in that case, chances are that most of them are false positives.</li><li>Sanity check based on expression: Do a quick calculation of average difference in expression for top DE genes between the two cohorts. If consistently you are observing a small difference in the average expression between the compared cohorts for most genes, chances are your DE results are full of false positives.</li></ol><h3>Ending Notes</h3><blockquote>Complexities should not be added unless necessary — Authors of Towards Data Science blog cited in the blog.</blockquote><p>As pointed out by authors in the cited studies, one shouldn’t assume that a simple method such as limma or Wilcoxon rank-sum test isn’t of use in front of advance differential expression methods. The blog also doesn’t exist to say that methods like DESeq2 and edgeR aren’t of use anymore. Simply, the choice of method should be dependent on data as the context. Instead of relying on the complexity of the method, one should be dependent on the expression data itself to decide the DE method for use.</p><p>I would like to end the blog with a question.</p><p>Do you think there will be additional problems with DE of even higher dimensional data such as single cell RNA sequencing ?</p><h3>References</h3><ol><li>Li, Yumei, et al. “<a href="https://doi.org/10.1186/s13059-022-02648-4">Exaggerated false positives by popular differential expression methods when analyzing human population samples.</a>” <em>Genome biology</em> 23.1 (2022): 1–13.</li><li><a href="https://towardsdatascience.com/deseq2-and-edger-should-no-longer-be-the-default-choice-for-large-sample-differential-gene-8fdf008deae9">https://towardsdatascience.com/deseq2-and-edger-should-no-longer-be-the-default-choice-for-large-sample-differential-gene-8fdf008deae9</a></li><li>Rocke, David M., et al. “<a href="https://www.biorxiv.org/content/10.1101/020784v1">Controlling false positive rates in methods for differential gene expression analysis using RNA-seq data.</a>” <em>Biorxiv</em> (2015): 018739.</li><li>Ritchie, Matthew E., et al. “<a href="https://doi.org/10.1093/nar/gkv007">limma powers differential expression analyses for RNA-sequencing and microarray studies.</a>” <em>Nucleic acids research</em> 43.7 (2015): e47-e47.</li><li>Love, Michael I., Wolfgang Huber, and Simon Anders. “<a href="http://Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2.">Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2.</a>” <em>Genome biology</em> 15.12 (2014): 1–21.</li><li>Robinson, Mark D., Davis J. McCarthy, and Gordon K. Smyth. “<a href="https://doi.org/10.1093/bioinformatics/btp616">edgeR: a Bioconductor package for differential expression analysis of digital gene expression data.</a>” <em>bioinformatics</em> 26.1 (2010): 139–140.</li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=379784ddee51" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>