Modeling PROTAC Degradation Activity with Machine Learning

PROTACs are a promising therapeutic modality that harnesses the cell'sbuilt-in degradation machinery to degrade specific proteins. Despite theirpotential, developing new PROTACs is challenging and requires significantdomain expertise, time, and cost. Meanwhile, machine learning has transformeddrug design and development. In this work, we present a strategy for curatingopen-source PROTAC data and an open-source deep learning tool for predictingthe degradation activity of novel PROTAC molecules. The curated datasetincorporates important information such as pDC_50, D_max, E3 ligasetype, POI amino acid sequence, and experimental cell type. Our modelarchitecture leverages learned embeddings from pretrained machine learningmodels, in particular for encoding protein sequences and cell type information.We assessed the quality of the curated data and the generalization ability ofour model architecture against new PROTACs and targets via three tailoredstudies, which we recommend other researchers to use in evaluating theirdegradation activity models. In each study, three models predict proteindegradation in a majority vote setting, reaching a top test accuracy of 82.6and 0.848 ROC AUC, and a test accuracy of 61generalizing to novel protein targets. Our results are not only comparable tostate-of-the-art models for protein degradation prediction, but also part of anopen-source implementation which is easily reproducible and lesscomputationally complex than existing approaches.