Uncovering biomarker genes with enriched classification potential from Hallmark gene sets

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Identifying reliable biomarkers from complex gene expression data is challenging.
• High-dimensional datasets require computationally efficient methods for biomarker discovery.
• Understanding gene expression patterns is crucial for linking genotype to phenotype.

Purpose of the Study:

• To develop and validate Gene Oracle, an algorithm for identifying salient biomarker genes.
• To evaluate the classification potential of gene sets and identify candidate biomarker genes.
• To explore the relationship between reduced functional complexity and improved biomarker potential.

Main Methods:

• Utilized a neural network within the Gene Oracle algorithm to assess polygenic classification potential of user-defined gene sets.
• Employed a combinatorial approach to decompose gene sets into candidate and non-candidate biomarker genes.
• Tested the algorithm on RNAseq data from The Cancer Genome Atlas (TCGA) and Genotype-Tissue Expression (GTEx) repositories, using Molecular Signatures Database (MSigDB) gene sets.

Main Results:

• Identified MSigDB Hallmark subsets with significant classification potential across both TCGA and GTEx datasets.
• Successfully pinpointed the most discriminatory candidate biomarker genes within each Hallmark gene set.
• Provided evidence suggesting that reduced functional complexity enhances the biomarker potential of identified genes.

Conclusions:

• Gene Oracle offers a computationally efficient method for identifying salient biomarker genes from large-scale gene expression data.
• The algorithm effectively leverages neural networks and combinatorial strategies for biomarker discovery.
• Findings suggest that simplifying functional complexity within gene sets can lead to more potent biomarkers.