Deep Sequence Modeling for Drug Classification: A Stacked LSTM-RNN Approach to WHO ATC Prediction Stacked

Abstract

The WHO Anatomical Therapeutic Chemical (ATC) classification system is a generally acknowledged drug classification scheme. The system is divided into five levels, each of which has multiple classes. Drugs were divided into classes based on their medicinal properties and effects. Predicting the drug's ATC code is critical in drug discovery and repurposing. For finding newer association data relating to ATC codes and drugs is still tough. Some compounds or drugs, in particular, may fall into 2 or more ATC classes. We presented a stacked LSTM-RNN model for ATC classification in order to resolve this concern. Thorough cross validation results have shown that staked LSTM achieved better results, compared to results achieved by the traditional ml algorithms which was represented in literature, specifically in AT (absolute true) rate, it is the most harsh and crucial metric for multi label systems.