Predicting PROTAC-targeted Degradation and Designing Androgen Receptor Degraders with AiPROTAC

ABSTRACT Proteolysis-targeting chimeras (PROTACs), a pioneering class of heterobifunctional ligands, have emerged as transformative tools in combating cancer and immune-related diseases due to their ability to target previously undruggable proteins and overcome drug resistance. Accurate assessment of the degradation potential of PROTACs is essential for advancing their therapeutic applications. However, the development of robust predictive models is hindered by the scarcity of large-scale datasets and domain-specific tools, as well as the underutilization of existing unlabeled data. To address these challenges, we developed an innovative method called AiPROTAC to predict the degradation capacity of PROTACs. In particular, AiPROTAC integrates graph augmentation, message-passing enhanced encoders and cross-attention mechanisms within a contrastive learning framework. Moreover, AiPROTAC leverages a curated specialized dataset combined with the PROTAC-DB dataset to enhance prediction accuracy. Experimental results across two datasets and six evaluation metrics demonstrate that AiPROTAC consistently outperforms state-of-the-art models. Further case studies underline its superior sensitivity and reliability. Notably, AiPROTAC facilitated the design of a novel androgen receptor (AR) degrader, PROTAC GT19, which achieved enhanced in vitro degradation compared to Bavdegalutamide (ARV-110). This advancement highlights our AiPROTAC’s potential to accelerate PROTAC-based therapeutic development and optimization.