Simultaneous regression and classification for drug sensitivity prediction using an advanced random forest method.

Area of Science:

• Computational biology
• Genomics
• Drug discovery

Background:

• Machine learning models predict anti-cancer drug efficacy using cancer cell line data.
• Existing methods face challenges with imbalanced drug response data, impacting sensitive cell line predictions.
• Class imbalance and regression imbalance negatively affect classification and regression performance, respectively.

Purpose of the Study:

• To introduce a novel approach, SimultAneoUs Regression and classificatiON Random Forests (SAURON-RF), for joint regression and classification analysis.
• To address the performance limitations caused by imbalanced drug response data in cancer cell line panels.
• To enhance the prediction accuracy for sensitive cell lines in anti-cancer therapy prediction.

Main Methods:

• Developed SAURON-RF, a novel machine learning algorithm performing simultaneous regression and classification.
• Trained and evaluated SAURON-RF on cancer cell line panels with imbalanced drug response data.
• Compared SAURON-RF performance against traditional regression and classification-only approaches.

Main Results:

• SAURON-RF significantly improved classification and regression performance for sensitive cell lines.
• A moderate performance decrease was observed for drug-resistant cell lines.
• Simultaneous regression and classification demonstrated superior predictive power compared to individual approaches.

Conclusions:

• SAURON-RF effectively handles imbalanced drug response data in cancer cell line prediction.
• Joint regression and classification analysis offers a more robust approach for predicting anti-cancer drug effectiveness.
• The SAURON-RF method advances the prediction of optimal anti-cancer therapies for a broader range of cell sensitivities.