1134-Reliable Drug Sensitivity Prediction and Prioritization

This research introduces a machine learning framework designed to improve the reliability and prioritization of anti-cancer drug treatments. Central to this approach is the use of conformal prediction, a mathematical method that provides certainty guarantees for drug sensitivity results rather than simple point estimates. To enhance clinical relevance, the authors developed CMax viability, a novel sensitivity measure that allows for the direct comparison and ranking of different drugs based on peak human plasma concentrations. This pipeline perform simultaneous classification and regression, effectively identifying the drugs which works and how efficient they will be. Experimental results using the GDSC database demonstrate that this system significantly reduces false predictions, achieving a median precision of 92% in identifying effective treatments. Ultimately, the work offers a more trustworthy decision-support tool for personalized oncology by quantifying prediction uncertainty in real-world medical scenarios.

References:

• Lenhof K, Eckhart L, Rolli LM, Volkamer A, Lenhof HP. Reliable anti-cancer drug sensitivity prediction and prioritization. Sci Rep. 2024 May 29;14(1):12303. doi: 10.1038/s41598-024-62956-6. PMID: 38811639; PMCID: PMC11137046.