Active learning framework of informative p53 cancer rescue mutants

Abstract

P53 mutants are closely related to humor tumors. Unfortunately, in vitro testing of all possible mutation combinations to determine their cancer rescue effects is infeasible due to time and expense. Therefore, it would be very desirable to have a computer model to run in silico experiments. In this paper, we propose a framework for active learning that can be used in any membership model active learning which does not consider predicted class as a criterion. Because the number of positive instances is much more than the number of negative instances. An active learning strategy is proposed to dynamically balance the number of positive and negative instances. As a result, we get a relatively balanced training set relative to both positive and negative instances, which leads to results showing good performance relative to high precision as well as high recall.