Development of an AI Model for PROTAC Drug Activity Prediction

Abstract

Recent advances in molecular biology have accelerated the development of therapeutic strategies targeting disease-associated proteins. Such targeted therapies are actively being explored in both academia and the pharmaceutical industry due to their superior efficacy and reduced side effects compared to conventional treatments. Most targeted therapeutics inhibit protein function by binding to the active site of the target protein. However, they face limitations when the target protein lacks a druggable binding pocket or forms large protein complex, making it difficult to achieve effective inhibition. Consequently, over 80% of human proteins are classified as 'undruggable targets,' highlighting the need for alternative strategies to overcome these limitations. Targeted protein degradation (TPD) has emerged as a promising strategy to address this undruggable target challenge. Unlike traditional pharmaceutical approaches, TPD utilizes the cellular protein degradation mechanism to eliminate the target protein itself. Among various TPD strategies, PROteolysis TArgeting Chimera (PROTAC) which leverages the ubiquitin-proteasome system has gained significant attention. PROTACs can induce protein degradation without strong binding affinity and offer selectivity through the choice of E3 ligase. However, their relatively large molecular weights and complex synthesis procedures pose challenges in terms of time and cost. In this study, we developed an artificial intelligence (AI) model to predict the degradation activity of PROTAC molecules. Public data from PROTAC-DB and PROTACpedia were collected for model training and performance evaluation. The proposed model demonstrated consistent performance improvements in the condition of unseen protein, unseen PROTAC, and even the combined unseen setting. Furthermore, the generalization ability of model was validated using an independent external test set. An ablation study confirmed the proposed model architecture contributes to performance enhancement.