Effects of Multi-Omics Characteristics on Identification of Driver Genes Using Machine Learning Algorithms

Area of Science:

• Genomics and Bioinformatics
• Cancer Research
• Computational Biology

Background:

• Cancer arises from complex genomic and epigenetic alterations.
• Identifying cancer driver genes is essential but challenging.
• Previous studies primarily focused on mutation data, neglecting multi-omics insights.

Purpose of the Study:

• To develop a framework for analyzing multi-omics data to identify cancer driver genes.
• To compare the efficacy of single-feature types versus combined features in driver gene detection.
• To enhance the understanding of cancer mechanisms through comprehensive driver gene identification.

Main Methods:

• Utilized a framework incorporating four machine learning algorithms.
• Analyzed pan-cancer data comprising 19,636 genes and 75 characteristics across four feature types.
• Employed Kullback-Leibler divergence for feature analysis against Cancer Gene Census (CGC) and non-CGC genes.

Main Results:

• Single-feature analysis indicated mutational features as the most effective.
• Combined analysis revealed that the top 45 features, including mutations and three other omics types, outperformed mutation-only features.
• The top 45 features demonstrated superior performance in detecting cancer driver genes.

Conclusions:

• A multi-omics approach significantly enhances cancer driver gene identification.
• The developed framework provides a more comprehensive understanding of cancer-driving mechanisms.
• This study expands upon existing methods for detecting cancer drivers by integrating diverse biological data.