Representing mutations for predicting cancer drug response.

Area of Science:

• Genomics
• Computational Biology
• Cancer Research

Background:

• Predicting cancer drug response is crucial for effective treatment.
• Current models often rely on binary gene mutation status, overlooking mutation severity.
• Not all mutations within a gene have equivalent biological or clinical impact.

Purpose of the Study:

• To develop and assess predictive models for cancer drug response using quantitative mutation scoring methods.
• To compare the performance of quantitative mutation scoring models against traditional binary mutation status models.
• To investigate the generalizability of quantitative mutation scoring to various cancer drugs and targets.

Main Methods:

• Utilized leading quantitative mutation scoring methods: VEST4, CADD (gene function impact), and CHASMplus (cancer driver likelihood).
• Constructed predictive models incorporating these quantitative mutation features.
• Evaluated model performance in predicting cellular responses to targeted therapies, including dabrafenib for BRAF-V600 mutations.

Main Results:

• Predictive models using quantitative mutation scores accurately captured cellular responses to dabrafenib, outperforming models based on binary mutation status.
• Performance improvements were observed across multiple targeted therapies, expanding genetic indications for inhibitors of PIK3CA, ERBB2, EGFR, PARP1, and ABL1.
• Incorporating quantitative mutation features enhanced both the predictive performance and mechanistic understanding of drug response.

Conclusions:

• Quantitative mutation scoring offers a more nuanced and effective approach to predicting cancer drug response compared to binary mutation status.
• This methodology improves the accuracy of predictive models and deepens the understanding of drug mechanisms in cancer.
• The developed models and methods have broad applicability for various targeted cancer therapies.