InDEP: an interpretable machine learning approach to predict cancer driver genes from multi-omics data.

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning in Oncology

Background:

• Identifying cancer driver genes is crucial for understanding cancer and developing targeted therapies.
• Current methods often rely on complex, difficult-to-interpret multi-omics data integration models.

Purpose of the Study:

• To introduce InDEP, an interpretable machine learning framework for identifying cancer driver genes.
• To address the complexity and lack of interpretability in existing driver gene identification methods.

Main Methods:

• InDEP utilizes cascade forests and a KernelSHAP module for interpretable analysis.
• The framework integrates multi-omics data to identify key features of driver genes.
• It enables interpretation at both gene and cancer-type levels.

Main Results:

• InDEP accurately identifies driver genes and discovers novel patterns contributing to their classification.
• The framework demonstrated superior accuracy compared to state-of-the-art methods.
• Mutational features, particularly substitution-type mutations, were identified as primary drivers.

Conclusions:

• InDEP provides an accurate and interpretable method for cancer driver gene identification.
• The framework refines the cancer driver gene catalog and identifies reliable candidate genes.
• This facilitates advancements in precision oncology and biomedical knowledge discovery.