Development of automatic descriptor extraction algorithms for ligand prediction.

点击翻译按钮获取中文摘要

This study tackled the problem for ligand prediction. The problem is to predict whether specified chemical compounds are interacted with specified target proteins. To achieve accurate prediction of ligands interacting with target proteins, effective descriptors that describe the properties of chemical compounds are crucial. Global models cannot deal with descriptors specific to each local prediction problem. Local models suffer from small sample size problems. This study introduced a concept of transfer learning to develop automatic descriptor extraction algorithms that solve the above two problems simultaneously.