"Heterophily in Graph Neural Networks for Single-cell RNA-sequencing Da" by Lian Duan

Date of Award

2-28-2025

Publication Type

Master Thesis

Degree Name

M.Sc.

Department

Computer Science

Keywords

cell type prediction, cell-cell communication, graph neural networks, heterophily, homophily, single-cell RNA sequencing

Supervisor

Luis Rueda

Rights

info:eu-repo/semantics/embargoedAccess

Abstract

Motivation: Graph Neural Networks (GNNs) excel at capturing complex dependencies in structured data, making them particularly valuable in computational biology. We investigate GNNs for single-cell RNA sequencing (scRNA-seq) data, where relationships can be homophilic (similar nodes connect) or heterophilic (dissimilar nodes connect). However, standard GNN models (GCN, GraphSAGE, GAT, and MixHop) often assume homophily, which is not always valid for biological networks. To address this, we explore H2GCN and Gated Bi-Kernel GNNs (GBK-GNN), both designed for heterophilic data. We compare these methods against traditional GNNs and an MLP across six diverse scRNA-seq datasets by generating cell-cell communication graphs from ligand-receptor pairs using LIANA.

Results: Our comparative analysis demonstrates that GBK-GNN consistently outperforms traditional homophily-assuming GNNs and H2GCN when dealing with highly heterophilous datasets. This underscores GBK-GNN’s robustness and adaptability to a variety of graph structures. Our findings highlight the importance of considering data-specific characteristics in GNN applications, demonstrating that heterophily focused methods can effectively decipher the complex patterns within scRNA-seq data. By integrating multi-omics data, including gene expression profiles and L-R interactions, we pave the way for more accurate and insightful analyses in computational biology, offering a more comprehensive understanding of cellular environments and interactions.

Download

Available for download on Wednesday, August 27, 2025

Share

COinS