Comprehensive assessment of computational algorithms in predicting cancer driver mutations

Area of Science:

• Oncology
• Computational Biology
• Genomics

Background:

• Cancer initiation and evolution are driven by functional somatic mutations, termed driver mutations.
• Identifying these driver mutations in tumor cells is essential for precision cancer medicine.
• Numerous computational algorithms exist to predict missense variant effects, but their comparative performance is unassessed.

Purpose of the Study:

• To comprehensively assess the performance of computational algorithms for predicting cancer driver mutations.
• To provide insights into best practices for computationally prioritizing mutation candidates.
• To guide the development of future predictive algorithms.

Main Methods:

• Construction of five benchmark datasets including protein 3D structures, OncoKB annotations, TP53 mutation effects, xenograft tumor formation, and in vitro viability assays.
• Evaluation of 33 distinct computational algorithms using these datasets.
• Comparison of cancer-specific algorithms against general-purpose algorithms.

Main Results:

• CHASM, CTAT-cancer, DEOGEN2, and PrimateAI demonstrated superior performance compared to other evaluated algorithms.
• Cancer-specific algorithms significantly outperformed general-purpose algorithms in predicting driver mutations.
• Benchmark datasets were created to facilitate rigorous algorithm evaluation.

Conclusions:

• This study offers a comprehensive performance assessment of algorithms for predicting cancer driver mutations.
• Cancer-specific algorithms are recommended for prioritizing mutation candidates in clinical settings.
• Findings will inform end-users and guide the development of improved predictive tools.