Gene-Expression Based Predictor for Drug Selection and Prioritization Using Learning-to-Rank

摘要

Accurate drug selection for cancer suffering patient is one of the major goal of precision medicine. Therapeutic method for drug selection is based on clinical trial. This method is extremely costly, time-consuming and not worth for aggressive cancer. Knowledge based diagnostic or therapeutic tools are commonly trusted to significantly increase the success rate of cancer specific drug selection. This research work enhances machine learning based Learning-to-Rank (LTR) framework for cancer specific drug selection and prioritization. This model is built to mainly achieve: 1. selection of the sensitive drugs from the pool of drugs for cell line data and 2. list out the sensitive drugs for each cell line based on number of concentration response point and area-under-curve. Each drug have the drug score based on the drug-cell interaction. This feature value is used in order to rank the drugs for each cell line. LTR algorithm first undergoes training to learn the drug scoring function and use it to predict drug sensitivity for new cell line. Also, vise-versa can be achieved with same approach. Performance of proposed model analyzed on the CCLE and CCTRP v2 dataset. Experimental result on benchmark datasets demonstrates that LTR based drug selection method significantly outperforms other state-of-art.