Functional random forest with applications in dose-response predictions.

Area of Science:

• Computational biology
• Pharmacogenomics
• Machine learning

Background:

• Personalized medicine requires accurate drug sensitivity prediction for individual tumors.
• Current methods predict single drug response metrics (e.g., IC50), failing to capture the full sensitivity profile.
• A complete dose-response curve is crucial for optimizing patient drug dosage.

Purpose of the Study:

• To develop a method for predicting the entire drug dose-response curve.
• To enhance existing Random Forests approaches for improved drug sensitivity prediction.
• To move beyond single-metric predictions towards a comprehensive functional profile.

Main Methods:

• Proposed an enhancement to Random Forests, termed Functional Random Forest (FRF).
• Developed functional regression trees with modified node costs for dose/response dependence.
• Incorporated dose-dependent expression data as functional predictors.

Main Results:

• FRF demonstrated higher accuracy in predicting dose-response curves compared to standard Random Forests on CCLE and GDSC datasets.
• FRF showed superior performance in predicting functional responses from functional predictors using the HMS-LINCS dataset.
• The framework effectively predicts the entire functional response profile, not just summary metrics.

Conclusions:

• Functional Random Forest provides an enhanced predictive framework for complete drug sensitivity profiles.
• This approach offers a more nuanced understanding of drug response than single-metric predictions.
• The method holds significant potential for advancing personalized medicine and optimizing cancer treatment strategies.