Deep reinforcement learning for personalized treatment recommendation

• DOI: 10.1002/sim.9491
• Publication Date: 2022-06-18T09:20:44Z
• Authors: Mingyang Liu; Xiaotong Shen; Wei Pan
• Keywords: Personalized Medicine; Supervised Learning; Learning to Rank

In precision medicine, the ultimate goal is to recommend most effective treatment an individual patient based on patient-specific molecular and clinical profiles, possibly high-dimensional. To advance cancer treatment, large-scale screenings of cell lines against chemical compounds have been performed help better understand relationship between genomic features drug response; existing machine learning approaches use exclusively supervised learning, including penalized regression recommender systems. However, it would be more efficient apply reinforcement sequentially learn as data accrue, selecting promising therapy for a given then collecting from corresponding data. this article, we propose novel personalized ranking system called Proximal Policy Optimization Ranking (PPORank), which ranks drugs their predicted effects per line (or patient) in framework deep (DRL). Modeled Markov decision process, proposed method learns suitable continuously over time. As proof-of-concept, conduct experiments two sets addition simulated The results demonstrate that DRL-based PPORank outperforms state-of-the-art competitors learning. Taken together, conclude methods DRL great potential medicine should further studied.