Understanding the genetic basis of a disease requires identifying not only individual disease-related genes but also the complex relationships between those genes, and Rice University’s Jake R. Patock, Baylor College of Medicine’s Rinki Ratnapriya, and Rice University’s Arko Burman have developed a new method to accomplish this. The research team presents a graphical approach to effectively identify gene clusters from RNA-seq data, revealing how genes work together in disease processes. Their technique builds a network representing the co-expression of genes, then uses advanced algorithms to map genes into a mathematical space in which related genes cluster, and finally applies clustering techniques to identify these groups. This innovative method shows consistent and robust results when applied to age-related macular degeneration data. Importantly, its design ensures easy applicability to a wide range of diseases, facilitating the discovery of underlying genetic mechanisms and potential therapeutic targets.
Integrating genomics, networks, and machine learning
This research collection explores the intersection of genomics, machine learning, and network analysis, with a particular focus on age-related macular degeneration and cancer. This collection highlights the growing trend of integrating diverse data types, such as transcriptomics and genomics, for a more comprehensive understanding of biological processes. Machine learning techniques such as clustering and deep learning are frequently used to analyze genomic data, identify patterns, and predict outcomes. Network analysis has emerged as a powerful approach to model biological systems, identify important genes and proteins, and understand their interactions through techniques such as coexpression network inference.
Several studies have specifically utilized single-cell RNA-seq, demonstrating the growing interest in understanding cellular heterogeneity. Researchers frequently use dimensionality reduction and clustering techniques to simplify and organize complex genomic data. The growing emphasis on explainable AI reflects a desire to create more interpretable and transparent machine learning models.
Gene embedding reveals disease-related clusters
Scientists have developed a new graph-based method to identify disease-associated gene clusters using RNA-Seq data and robust clustering analysis. Research begins by constructing gene co-expression networks that represent relationships between genes based on expression patterns. The researchers then leveraged the Node2Vec+ algorithm to compute gene embeddings and create a numerical representation of each gene in the network to capture its contextual relationships with other genes. These embeddings represent genes in a multidimensional space and facilitate the identification of functionally similar genes.
Following the creation of the gene embeddings, the team implemented spectral clustering to identify distinct gene groups. This technique divides genes into clusters based on the distance between gene embeddings, revealing potential functional relationships and shared biological pathways. To ensure stability and optimality of the entire process, the researchers jointly used the tree-structured Parzen Estimator to optimize every step and tune parameters to maximize the reliability of the resulting clusters. The work begins by building a gene co-expression network and mapping the relationships between genes, then computing gene embeddings to represent each gene as a point in a high-dimensional space that captures its expression pattern and connections within the network. Finally, the team clustered these gene embeddings to reveal groups of genes with similar expression profiles, potentially indicating common functions or pathways involved in the disease. The core breakthrough lies in a joint optimization process that simultaneously refines network construction, embedding computation, and clustering to maximize overall performance.
This is in contrast to traditional approaches that optimize each step individually. The team cost function focuses on the final clustering quality and ensures good performance. The research team constructed a gene co-expression network from RNA-seq data and used the advanced algorithm Node2Vec+ to compute gene embeddings to effectively represent genes within the network context. Subsequent spectral clustering of these embeddings successfully identified gene groups and revealed potential functional relationships. We demonstrate that this method can consistently generate robust gene clusters from RNA-Seq data, providing a valuable tool for understanding complex disease mechanisms. By grouping known AMD-associated genes, this approach helps pinpoint common biological pathways and uncover previously unrecognized genes involved in the disease. This clustering strategy aims to go beyond focusing on individual genes and identify underlying disease mechanisms for more effective therapeutic interventions.
