<?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 Ray Hotate on Medium]]></title>
        <description><![CDATA[Stories by Ray Hotate on Medium]]></description>
        <link>https://medium.com/@rayhtt?source=rss-339665d7c635------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*lKIhV1UbzCyyoxtM</url>
            <title>Stories by Ray Hotate on Medium</title>
            <link>https://medium.com/@rayhtt?source=rss-339665d7c635------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 01:04:59 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@rayhtt/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[Hierarchical Link Prediction for Drug-Drug Interactions]]></title>
            <link>https://medium.com/stanford-cs224w/hierarchical-link-prediction-for-drug-drug-interactions-0128403c0b02?source=rss-339665d7c635------2</link>
            <guid isPermaLink="false">https://medium.com/p/0128403c0b02</guid>
            <dc:creator><![CDATA[Ray Hotate]]></dc:creator>
            <pubDate>Fri, 03 Jan 2025 19:48:18 GMT</pubDate>
            <atom:updated>2025-01-03T19:48:18.216Z</atom:updated>
            <content:encoded><![CDATA[<p>Luke Park, Irfan Nafi, Ray Hotate</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/0*NBjvAfSmDUjfFNL5" /><figcaption>Effects of DDI</figcaption></figure><h3>1. Introduction</h3><p>Graph Machine Learning (Graph ML) has emerged as a transformative approach to solving complex real-world problems by leveraging the inherent structure of graph-based data. In drug discovery and pharmacology, this approach is especially promising, where molecular, biological, and interaction networks can be modeled as graphs to extract meaningful insights. Predicting drug-drug interactions (DDIs) is a critical challenge in healthcare, as interactions between medications can lead to unforeseen and potentially severe adverse effects. Early identification of these interactions is crucial for ensuring patient safety and optimizing drug efficacy.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/550/0*67n8jccTT6qvXYH7" /><figcaption>Example of link prediction</figcaption></figure><p>Traditional approaches to DDI prediction often rely on pharmacokinetic models or large-scale clinical data analysis, which are time-consuming and limited in generalizability (Cami et al., 2013). More recently, graph-based machine learning methods, including Graph Neural Networks (GNNs), have shown the ability to predict DDIs effectively by leveraging the topological structure of interaction networks (Sun et al., 2019). These methods provide a scalable and efficient alternative to traditional techniques, addressing the limitations of insufficient clinical evidence and high-dimensional data representation.</p><p>Our study introduces a hierarchical approach using Graph Attention Networks (GATs) to enhance DDI prediction. Unlike previous methods, our hierarchical model combines local embeddings, which capture immediate interaction patterns, with community-level embeddings, which represent broader network structures. This dual-level approach allows us to incorporate both fine-grained and global perspectives, aiming to improve the robustness and accuracy of predictions.</p><h3><strong>2. Motivation</strong></h3><p>Drug interactions can exhibit synergistic or antagonistic effects that significantly alter therapeutic outcomes. Despite substantial advances in computational techniques, many prediction models fail to generalize to unseen data, particularly when tested on drugs with novel mechanisms of action or targeting new proteins. The ogbl-ddi dataset, which represents a comprehensive drug-drug interaction network, provides a unique opportunity to explore advanced modeling techniques in this domain. Its challenging protein-target split ensures that test data involve drugs interacting with different proteins than those in the training data, simulating real-world scenarios where drugs with novel targets emerge (Min &amp; Bae, 2017).</p><p>Our hierarchical GAT model addresses these challenges by leveraging attention mechanisms to focus on relevant portions of the graph at both local and community levels. This design not only enhances interpretability but also offers insights into how drugs with disparate biological mechanisms may interact, thus contributing to a safer drug development pipeline.</p><h3>3. Dataset</h3><p>For this project, we utilized the <strong>ogbl-ddi</strong> dataset, a well-curated graph dataset representing the drug-drug interaction network. This dataset comprises <strong>nodes corresponding to FDA-approved or experimental drugs</strong> and <strong>edges representing interactions</strong> between these drugs. These interactions highlight cases where the combined effect of two drugs significantly deviates from their independent actions.</p><p>The <strong>ogbl-ddi</strong> dataset is particularly well-suited for this study due to its <strong>homogeneous, unweighted, and undirected graph structure</strong>. The primary task is to predict interactions between drugs by leveraging known interactions, with the objective of ranking true interactions higher than negative samples.</p><p>To evaluate the practical utility of our approach, we implemented a <strong>protein-target split</strong> for dataset splitting. This method ensures that the <strong>test set includes drugs targeting proteins significantly different</strong> from those in the training and validation sets. By adopting this split, we assess our model’s ability to predict interactions for drugs with novel biological mechanisms. Such capability is crucial for identifying <strong>breakthrough discoveries in pharmacology</strong> and paving the way for innovative treatments.</p><h3>Graph Statistics</h3><ul><li><strong>Number of nodes:</strong> 4,267</li><li><strong>Number of edges:</strong> 2,135,822</li></ul><h4>Split Sizes:</h4><ul><li><strong>Training edges:</strong> 1,067,911</li><li><strong>Validation edges:</strong> 133,489</li><li><strong>Test edges:</strong> 133,489</li></ul><h4>Edge Split Ratio:</h4><ul><li><strong>Train:</strong> 80.00%</li><li><strong>Valid:</strong> 10.00%</li><li><strong>Test:</strong> 10.00%</li></ul><h3><strong>4. Prior Work</strong></h3><p>The paper “Can GNNs Learn Link Heuristics” by Shuming Liang et al. investigates the effectiveness of GNNs in learning structural information, specifically in the case of link prediction tasks. Although structural identification is not the primary task of link prediction, encapsulating such information is important and would help performance. Their paper questions the ability of GNN aggregation methods in capturing more nuanced and complex structural information and attempts to alleviate this problem by explicitly passing in features that measure certain structural properties.</p><p>One of the key heuristics in Liang’s paper is the <strong>common neighbors metric</strong>. The common neighbors metric looks at the sets of neighbors between two entities and if the sets have a non-zero intersection then the two nodes have a common neighbors. In their paper, they run several ablations studies via mean aggregation and graph attention aggregation methods to show that standard GNNs cannot accurately determine whether or not two nodes share a common neighbor. The importance of the common neighbor is further amplified by the fact that drug interactions often occur in multi-hop relations.</p><p>The authors further emphasize that the inability of standard GNN aggregation schemes to effectively capture the number of common neighbors stems from the <strong>inherent limitations of set-based pooling operations</strong>. These aggregation methods, such as mean pooling and attention mechanisms, treat the node neighborhood as an unordered set. Consequently, they fail to encode structural properties such as the size of the intersection of neighborhoods or the presence of specific patterns like common neighbors. Liang et al. argue that this shortcoming significantly hampers the performance of GNN-based link prediction models, particularly in applications where multi-hop relations play a critical role, such as drug interaction networks.</p><p>To address this limitation, the paper proposes the incorporation of explicit structural features into GNN architectures. These features are derived from traditional link prediction heuristics, such as the Common Neighbors (CN), Adamic-Adar (AA) index, and Jaccard coefficient, which quantify various aspects of the structural relationship between node pairs. By encoding these metrics as additional features or trainable embeddings, Liang et al. demonstrate that GNN-based models can better capture structural dependencies and improve link prediction performance.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/220/0*68oR6aES6Vb-ZJu0" /><figcaption>Community level structures communicating with each other</figcaption></figure><p>In their experimental framework, Liang et al. evaluate the impact of integrating these structural features across multiple datasets with varying graph densities. Their results highlight a strong positive correlation between graph density and the benefits of incorporating structural heuristics. Specifically, on dense graphs, the inclusion of structural features, such as CN, led to significant performance improvements, validating the hypothesis that explicit structural encoding compensates for the limitations of GNN aggregation schemes.</p><p>Additionally, the paper critiques existing methods such as SEAL and NBFNet, which are popular in the domain of link prediction. SEAL-type models rely on the classification of enclosing subgraphs and assume that GNNs can learn structural heuristics like CN implicitly. However, the findings in Liang et al.’s paper reveal that these models fail to encode structural information effectively due to the same aggregation limitations observed in standard GNNs. Similarly, NBFNet’s edge-centric approach lacks the capacity to train powerful node embeddings, limiting its performance on dense graphs.</p><p>The insights from Liang et al.’s work provide a strong argument for hybrid approaches that combine the strengths of traditional heuristics with the representational power of GNNs. Their findings underscore the importance of explicitly incorporating structural features into GNN-based link prediction frameworks, paving the way for more robust and interpretable models that can operate effectively across a wide range of graph structures.</p><h3><strong>5. Methods</strong></h3><p>Building on the foundational insights from the work of Liang et al. in <em>“Can GNNs Learn Link Heuristics?”</em>, our team sought to address the limitations of standard GNNs in capturing nuanced structural information critical for link prediction tasks. Liang et al.’s work emphasized the challenges faced by GNN aggregation methods in encoding pair-specific heuristics such as the number of common neighbors, highlighting the inherent limitations of set-based pooling operations. Inspired by their findings and the need to improve GNN performance on tasks requiring complex structural reasoning, we extended this line of research by introducing a novel <strong>hierarchical attention-based graph neural network architecture</strong>.</p><pre>class HC-GAM(Predictor):<br>    def __init__(self, args):<br>        super().__init__(args)<br>        <br>        # Get the actual dimension being used (from parent class logic)<br>        dim_hidden = args.dim_in if args.dim_hidden is None else args.dim_hidden<br>        <br>        # Additional components for hierarchical model<br>        self.text_encoder = nn.Linear(args.text_dim, args.dim_encoding)<br>        self.relation_encoder = nn.Linear(args.relation_dim, args.dim_encoding)<br>        <br>        # Local and Global Attention (in addition to existing ComHG layers)<br>        self.local_attention = nn.ModuleList([<br>            MultiHeadAttention(args.dim_encoding, args.n_heads) <br>            for _ in range(args.n_layers)<br>        ])<br>        <br>        self.global_attention = nn.ModuleList([<br>            MultiHeadAttention(args.dim_encoding, args.n_heads)<br>            for _ in range(args.n_layers)<br>        ])<br>        <br>        # Community-based components<br>        self.community_encoder = nn.Linear(args.num_communities, args.dim_encoding)<br>        <br>        # Additional MLP for combining hierarchical features<br>        self.hierarchical_mlp = nn.Sequential(<br>            nn.Linear(dim_hidden * 3, dim_hidden * 2),<br>            nn.ReLU(),<br>            nn.Dropout(args.dropout),<br>            nn.Linear(dim_hidden * 2, dim_hidden)<br>        )<br><br>    def forward(self, graph, edge_batch):<br>        # Original ComHG processing<br>        x_original = super().forward(graph, edge_batch)<br>        <br>        # Additional hierarchical processing<br>        if hasattr(graph, &#39;text_features&#39;) and hasattr(graph, &#39;communities&#39;):<br>            # Text embedding processing<br>            text_emb = self.text_encoder(graph.text_features)<br>            <br>            # Local attention<br>            local_features = text_emb<br>            for layer in self.local_attention:<br>                local_features = layer(local_features, graph.adj)<br>            <br>            # Global community-based attention<br>            community_emb = scatter_mean(text_emb, graph.communities, dim=0)<br>            global_features = text_emb<br>            for layer in self.global_attention:<br>                global_features = layer(global_features, community_emb)<br>            <br>            # Community encoding<br>            community_features = self.community_encoder(<br>                F.one_hot(graph.communities, num_classes=self.args.num_communities).float()<br>            )<br>            <br>            # Combine features<br>            combined_features = torch.cat([<br>                x_original,<br>                local_features[edge_batch[:, 0]] * local_features[edge_batch[:, 1]],<br>                global_features[edge_batch[:, 0]] * global_features[edge_batch[:, 1]]<br>            ], dim=-1)<br>            <br>            # Final prediction combining original and hierarchical features<br>            x = self.hierarchical_mlp(combined_features)<br>            return torch.sigmoid(x)<br>        <br>        # Fallback to original prediction if hierarchical features aren&#39;t available<br>        return x_original</pre><p>Our model integrates hierarchical and community-based attention mechanisms to enhance structural representation learning. While Liang et al. proposed incorporating explicit structural heuristics, we extended this approach by embedding these heuristics within a broader framework of hierarchical feature interactions. Our architecture, termed the <strong>Hierarchical Community-based Graph Attention Model (HC-GAM)</strong>, was specifically designed to leverage both local and global graph structures through multi-scale attention mechanisms.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Xo5XrcUUdHyX7teMcyYc0g.png" /></figure><p>A key innovation in our work lies in the use of <strong>hierarchical attention mechanisms</strong> to model interactions at multiple levels of graph organization. We incorporated community structures derived from algorithms such as the Leiden method, embedding these structural roles into the graph’s representation. Two distinct attention layers — local and global — were employed to capture fine-grained node interactions within communities and broader relational patterns across them. Local attention mechanisms focused on preserving the detailed structural relationships within neighborhoods, while global attention mechanisms captured the influence of higher-order communities, providing a comprehensive hierarchical understanding of the graph.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*J2c6C5T9PKBue27ji2ygvQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WS0eXYCJbsFVF978d_mS5g.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GCgb0GpThFAqxoVQQv9e8w.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*f8uBtiMkdZUL5IXDbSwazQ.png" /></figure><p>In addition to hierarchical attention, we introduced a <strong>multi-level feature fusion process</strong> that synthesizes local, global, and community-based features. This fusion was achieved through a hierarchical multi-layer perceptron (MLP), which combined original graph embeddings from extended ComHG layers, node interactions derived through local attention mechanisms, and global community-driven interactions. By integrating these components, the model effectively balances the importance of granular local features and larger-scale structural patterns, addressing a critical gap in traditional GNN designs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YOsrgNRIELMikSltdb44gg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LWAxfIqCfca5Ba_1Gz2vdA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-H27gyqc3QNEo_W9DWv16g.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*y5Zp6R4bXyG44CKjtHSsTw.png" /></figure><p>To further enhance structural learning, we adopted and extended explicit structural heuristics such as Common Neighbors (CN), Adamic-Adar, and Jaccard coefficients. These heuristics were encoded as additional trainable embeddings, complementing the learned features generated by the GNN. This integration of heuristic-based and learned features allowed the model to better capture relational properties that are vital for link prediction, particularly in dense and complex graphs.</p><p>Our architecture also incorporated <strong>multi-head attention mechanisms</strong> that enabled flexible and adaptive processing of node and adjacency matrix features. By employing combinatorial strategies such as concatenation and multiplication, the attention layers dynamically adapted to the graph’s structural characteristics, further improving the model’s ability to represent intricate graph relationships. This innovation, coupled with the hierarchical modeling approach, provided a significant improvement in the model’s capacity to represent long-range dependencies and higher-order graph interactions.</p><pre>class MultiHeadAttention(nn.Module):<br>    def __init__(self, dim_in, n_heads):<br>        super().__init__()<br>        self.n_heads = n_heads<br>        self.dim_k = dim_in // n_heads<br>        <br>        self.Q = nn.Linear(dim_in, dim_in)<br>        self.K = nn.Linear(dim_in, dim_in)<br>        self.V = nn.Linear(dim_in, dim_in)<br>        <br>        self.out = nn.Linear(dim_in, dim_in)<br>        <br>    def forward(self, x, adj, mask=None):<br>        B = x.size(0)<br>        N = x.size(0) if len(x.size()) == 2 else x.size(1)<br>        <br>        # Reshape for multi-head attention<br>        Q = self.Q(x).view(B, N, self.n_heads, self.dim_k)<br>        K = self.K(x).view(B, N, self.n_heads, self.dim_k)<br>        V = self.V(x).view(B, N, self.n_heads, self.dim_k)<br>        <br>        # Scaled dot-product attention<br>        scores = torch.matmul(Q, K.transpose(-2, -1)) / math.sqrt(self.dim_k)<br>        <br>        if mask is not None:<br>            scores = scores.masked_fill(mask == 0, -1e9)<br>        <br>        # Apply attention<br>        attention = F.softmax(scores, dim=-1)<br>        out = torch.matmul(attention, V)<br>        <br>        # Reshape and project output<br>        out = out.reshape(B, N, -1)<br>        return self.out(out)</pre><p>Training and testing were done using the following functions. The GPU we leveraged was NVIDIA L40S with 64G RAM.</p><pre>def train(args, predictor, optimizer, scheduler, data_pos, data_neg):<br>    predictor.train()<br>    running_loss = running_examples = 0<br><br>    pos_edges, pos_perms = DataLoader(args, data_pos)<br>    neg_edges, neg_perms = DataLoader(args, data_neg)<br>    leniter = min(len(pos_perms), len(neg_perms))<br>    for i in range(leniter):<br>        optimizer.zero_grad()<br>        edge_pos = pos_edges[pos_perms[i]]<br>        edge_neg = neg_edges[neg_perms[i]]<br>        y_pos = predictor(data_pos, edge_pos)<br>        y_neg = predictor(data_neg, edge_neg)<br>        <br>        if len(y_pos) != len(y_neg):<br>            break<br><br>        pos_loss = (-torch.log(y_pos + 1e-15)).mean()<br>        neg_loss = (-torch.log(1.0 - y_neg + 1e-15)).mean()<br>        loss = pos_loss + neg_loss<br>        loss.backward()<br>        torch.nn.utils.clip_grad_norm_(predictor.parameters(), args.clip_grad_norm)<br>        optimizer.step()<br><br>        running_loss += loss.item()<br>        running_examples += 1<br>    return running_loss / running_examples</pre><pre>@torch.no_grad()<br>def test(args, predictor, data_pos, data_neg):<br>    predictor.eval()<br><br>    def get_predictions(args, predictor, data_pred):<br>        pred = []<br>        edges = data_pred.edges<br>        num_edge = edges.size(0)<br>        step, end = args.batch_size * 100, num_edge<br>        perms = [np.array(range(i, i + step)) if i + step &lt; end else np.array(range(i, end)) for i in range(0, end, step)]<br>        for perm in perms:<br>            edge_batch = edges[perm]<br>            y = predictor(data_pred, edge_batch)<br>            pred.append(y.detach().cpu().squeeze(1))<br>        return torch.cat(pred, dim=0)<br><br>    pred = torch.cat([get_predictions(args, predictor,data_pos), get_predictions(args, predictor,data_neg)], dim=0)<br>    true = torch.cat([torch.ones(data_pos.edges.size(0)), torch.zeros(data_neg.edges.size(0))], dim=0)<br><br>    return pred, true</pre><p>Our HC-GAM demonstrated robust performance, significantly outperforming baseline methods on Hits@20. The inclusion of hierarchical attention mechanisms and multi-scale feature fusion proved instrumental in achieving these results. Ablation studies confirmed the complementary contributions of local and global attention mechanisms, as well as the critical role of heuristic embedding in enhancing predictive accuracy. The full code for the execution of the model is accessible in the <a href="https://colab.research.google.com/drive/1vLC1UzorokCuDjH8gV9bCwssCpa-W3f8?usp=sharing">Google Colab</a>.</p><h3><strong>6. Results</strong></h3><p>The model was trained over 1,000 epochs, achieving an final test accuracy of <strong>96.69%</strong>, outperforming Liang et al.’s model, which reported a test accuracy of <strong>95.49%</strong>. Utilizing <strong>8,896,994 parameters</strong>, the model showcased efficiency and precision. For comparison, the state-of-the-art HyperFusion model employs nearly <strong>1 billion parameters</strong>, while the second and third-ranking models leverage approximately <strong>10 million</strong> and <strong>5 million parameters</strong>, respectively, highlighting the balance of scalability and performance in our approach.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*I4WfUHTMbDZxRoV541EoIw.png" /><figcaption>Leaderboard for ogbl-ddi dataset</figcaption></figure><p>Our model performs better than Graph Inception Diffusion Networks (GIDN) by Wang et al., which generalizes graph diffusion across multiple feature spaces a incorporates an inception module to reduce the computational overhead caused by complex network structures. However, we perform worse than Hyperfusion by Zhang et al. and Ensemble Learning for GNNs (ELGNN) by Wong et al., which both deploy ensembles of GNNS, with Hyperfusion using a hypergraph to “fuse” outputs from various GNNs. These ensembles have many more parameters and have more diverse exploration of features than we are capable of with a single instance of our hierarchical GAT, which may explain increased accuracy.</p><p>The graph below compares the test error rates of the HC-GAM model and Liang et al.’s model over epochs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1005/1*0TnYmJTFoM-yjnQbqXVYAA.png" /></figure><p>Both models consistently achieved over 90% accuracy after the 600th epoch. To facilitate a more detailed comparison, we visualized the log-scaled error rates below.</p><p>Initially, the HC-GAM model exhibited a higher error rate, indicating lower accuracy compared to Liang et al.’s model. However, after approximately the 850th epoch, HC-GAM consistently <strong>achieved superior accuracy</strong>. To address the high variance in test accuracy observed between epochs 200–600 for both models, a moving average with a window size of 10 epochs is shown in red and green for clarity. The translucent dots in blue and orange shows the test accuracy for each epoch.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1012/1*gNquwLp9wf-XtZidAWpuXg.png" /></figure><p>The graph highlights the effectiveness of our hierarchical attention mechanism. Additionally, given that our model utilized 1.6 times more parameters than Liang et al.’s model, it is reasonable that it required more epochs to achieve superior accuracy.</p><h3><strong>7. Future Work</strong></h3><p>The field of drug-drug interaction (DDI) prediction holds immense potential for further advancements, particularly through the integration of emerging techniques and technologies. One promising direction involves the application of pre-trained Large Language Models (LLMs) for link prediction tasks. As demonstrated in approaches like GraphGPT (Zhao et al., 2013), Next Token Prediction (NTP) can effectively encode graph structures into sequences of tokens. By leveraging such pre-trained LLMs, future research can explore their ability to predict new edges in heterogeneous graphs, potentially enhancing the capacity to capture complex relationships in networks such as those representing DDIs.</p><p>Another critical avenue for exploration is the incorporation of multimodal data. Current graph-based models can benefit from integrating diverse data types, including chemical structures, clinical trial results, and pharmacological properties, to develop more informative node and edge representations. By combining graph embeddings with text or image embeddings from LLMs, models may achieve improved generalization across diverse datasets, enabling a broader range of applications in pharmacology.</p><p>Scalability and computational efficiency are also key challenges that must be addressed to extend the hierarchical GAT approach to larger-scale datasets. Implementing techniques such as subgraph sampling and scaling mechanisms, as proposed in transformer-based graph models, can allow for real-time predictions on extensive biological networks without compromising performance. In parallel, future efforts should prioritize improving biological interpretability. Beyond predictive accuracy, it is important to link model predictions to specific biological pathways or mechanisms of action, providing insights that can inform drug safety and development.</p><p><strong>Reference</strong></p><p>Liang, S., Ding, Y., Li, Z., Liang, B., Zhang, S., Wang, Y., &amp; Chen, F. (2024). Can GNNs Learn Link Heuristics? A Concise Review and Evaluation of Link Prediction Methods. <em>arXiv preprint arXiv:2411.14711</em>.</p><p>Cami, A., Manzi, S., Arnold, A., &amp; Reis, B. Y. (2013). Pharmacointeraction network models predict unknown drug-drug interactions. PLoS ONE, 8(4), e61468. <a href="https://doi.org/10.1371/journal.pone.0061468">https://doi.org/10.1371/journal.pone.0061468</a></p><p>Cascorbi, I. (2012). Drug interactions. Deutsches Ärzteblatt International. <a href="https://doi.org/10.3238/arztebl.2012.0546">https://doi.org/10.3238/arztebl.2012.0546</a></p><p>Link property prediction. (2024, December 4). Open Graph Benchmark. <a href="https://snap-stanford.github.io/ogb-web/docs/linkprop/">https://snap-stanford.github.io/ogb-web/docs/linkprop/</a></p><p>Min, J. S., &amp; Bae, S. K. (2017). Prediction of drug–drug interaction potential using physiologically based pharmacokinetic modeling. Archives of Pharmacal Research, 40(12), 1356–1379. <a href="https://doi.org/10.1007/s12272-017-0976-0">https://doi.org/10.1007/s12272-017-0976-0</a></p><p>Sun, M., Zhao, S., Gilvary, C., Elemento, O., Zhou, J., &amp; Wang, F. (2019). Graph convolutional networks for computational drug development and discovery. Briefings in bioinformatics. <a href="https://doi.org/10.1093/bib/bbz042">https://doi.org/10.1093/bib/bbz042</a>.</p><p>Wang, Z., Guo, Y., Zhao, J., Zhang, Y., Yu, H., Liao, X., Wang, B., &amp; Yu, T. (2024). Gidn: A lightweight graph inception diffusion network for high-efficient link prediction (arXiv:2210.01301). arXiv. <a href="https://doi.org/10.48550/arXiv.2210.01301">https://doi.org/10.48550/arXiv.2210.01301</a></p><p>Wong, Z. H., Yue, L., &amp; Yao, Q. (2023). Ensemble learning for graph neural networks (arXiv:2310.14166). arXiv. <a href="https://doi.org/10.48550/arXiv.2310.14166">https://doi.org/10.48550/arXiv.2310.14166</a></p><p>Zhang, X. (n.d.). HyperFusion/Multi_Model_Ensemble_on_Hypergraph.pdf at master · zhangxwww/HyperFusion. GitHub. Retrieved December 12, 2024, from <a href="https://github.com/zhangxwww/HyperFusion/blob/master/Multi_Model_Ensemble_on_Hypergraph.pdf">https://github.com/zhangxwww/HyperFusion/blob/master/Multi_Model_Ensemble_on_Hypergraph.pdf</a></p><p>Zhao, Q., Ren, W., Li, T., Xu, X., &amp; Liu, H. (2023). Graphgpt: Graph learning with generative pre-trained transformers (arXiv:2401.00529). arXiv. <a href="https://doi.org/10.48550/arXiv.2401.00529">https://doi.org/10.48550/arXiv.2401.00529</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0128403c0b02" width="1" height="1" alt=""><hr><p><a href="https://medium.com/stanford-cs224w/hierarchical-link-prediction-for-drug-drug-interactions-0128403c0b02">Hierarchical Link Prediction for Drug-Drug Interactions</a> was originally published in <a href="https://medium.com/stanford-cs224w">Stanford CS224W: Machine Learning with Graphs</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Test]]></title>
            <link>https://medium.com/@rayhtt/test-e853524f699f?source=rss-339665d7c635------2</link>
            <guid isPermaLink="false">https://medium.com/p/e853524f699f</guid>
            <dc:creator><![CDATA[Ray Hotate]]></dc:creator>
            <pubDate>Thu, 12 Dec 2024 22:02:53 GMT</pubDate>
            <atom:updated>2024-12-12T22:02:53.527Z</atom:updated>
            <content:encoded><![CDATA[<img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e853524f699f" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>