Network Biology-Inspired Machine Learning Features Predict Cancer Gene Targets and Reveal Target Coordinating...

Area of Science:

• Computational biology
• Genomics
• Translational oncology

Background:

• Understanding cancer's reliance on specific gene activities is crucial for developing new therapies.
• The Cancer Dependency Map (DepMap) project provides large-scale genetic screening data for cancer cell lines.

Purpose of the Study:

• To develop machine learning algorithms predicting cancer gene dependencies using network biology.
• To identify network features that coordinate these gene dependencies across different cancer types.

Main Methods:

• Utilized DepMap data for cancer gene dependency screening.
• Engineered novel machine learning features integrating network topology and biological annotations.
• Applied machine learning models to predict binary gene dependencies.

Main Results:

• Achieved high accuracy (F1 scores > 0.90) in predicting gene dependencies across all examined cancer types.
• Demonstrated robust model performance under various hyperparameter settings.
• Identified tumor-specific coordinators of gene dependency, such as gene connectivity in kidney and thyroid cancers, and cell death pathway associations in lung cancer.

Conclusions:

• Biologically informed network features enhance predictive pharmacology models for cancer.
• This approach provides valuable mechanistic insights into tumor-specific gene dependencies.
• The findings support the development of targeted cancer therapeutics based on predicted gene dependencies.