Reliable anti-cancer drug sensitivity prediction and prioritization.

Area of Science:

• Computational Biology
• Machine Learning Applications
• Pharmacogenomics

Background:

• Machine learning (ML) offers significant potential for real-world problem-solving but carries inherent risks.
• Ensuring the reliability of ML predictions, including minimizing model error and estimating uncertainty, is crucial for risk mitigation, especially in medical applications.
• Accurate anti-cancer drug sensitivity prediction is vital for effective cancer treatment and drug development.

Purpose of the Study:

• To develop a novel machine learning approach for reliable anti-cancer drug sensitivity prediction and prioritization.
• To guarantee user-specified certainty levels in ML predictions for clinical applications.
• To introduce a new drug sensitivity measure for straightforward drug prioritization based on clinical relevance.

Main Methods:

• A novel conformal prediction approach was developed and applied to classification, regression, and simultaneous regression/classification tasks.
• The method ensures user-specified prediction certainty levels, addressing a key challenge in ML reliability.
• A new drug sensitivity measure was formulated using clinically relevant drug concentrations.

Main Results:

• The developed approach provides reliable anti-cancer drug sensitivity predictions with guaranteed certainty.
• The conformal prediction framework is versatile, applicable across different ML task types.
• The novel drug sensitivity measure facilitates efficient and clinically relevant drug prioritization for individual cancer samples.

Conclusions:

• The presented approach enhances the reliability of ML predictions in anti-cancer drug sensitivity analysis.
• Guaranteed certainty levels in ML predictions are critical for safe and effective clinical decision-making.
• The novel drug sensitivity measure and prediction method offer a promising tool for personalized cancer therapy and drug discovery.