DegradeQuery: Counterfactual Tuple Pretraining for Context-Aware PROTAC Degradation Prediction

2026-08-11

AI Summary

This work addresses the challenge that PROTAC degrader activity is highly dependent on the ternary context formed by the PROTAC molecule, target protein, and E3 ligase, yet existing databases often lack degradation labels, limiting supervised learning approaches. To overcome this, the authors propose DegradeQuery, a novel framework that leverages unlabeled ternary complexes to construct counterfactual contrastive pretraining signals, enabling context-aware representation learning without relying on pseudo-labels. The model is subsequently fine-tuned on complete ternary complexes to predict degradation efficacy. By effectively extracting implicit supervisory signals from incompletely annotated data, DegradeQuery seamlessly integrates with protein language models. Evaluated on the PROTAC-8K benchmark, it achieves a ROC-AUC of 0.9065 and an accuracy of 0.8500, significantly outperforming current methods and demonstrating the effectiveness of unlabeled data–driven modeling in PROTAC activity prediction.

Abstract

Proteolysis-targeting chimeras (PROTACs) induce protein degradation by recruiting a target protein to an E3 ubiquitin ligase, making degradation a joint outcome of the degrader molecule and its biological context. Although public databases contain thousands of structured molecule-target-E3 records, degradation measurements are available for only a small fraction of them. Existing supervised approaches therefore leave most recorded chemical-biological relationships unused. We introduce DegradeQuery, a context-aware prediction framework that converts these label-missing records into a pretraining signal. Its counterfactual tuple pretraining objective contrasts recorded tuples with alternatives formed by replacing the target, the E3 ligase, or both, enabling the model to learn contextual associations without assigning activity pseudo-labels. The resulting representation is then fine-tuned to predict degradation from the complete molecule-target-E3 context. On the official PROTAC-8K benchmark, DegradeQuery achieves an area under the receiver operating characteristic curve of 0.9065 and an accuracy of 0.8500, outperforming the compared methods. Controlled analyses further show that the improvement is primarily attributable to tuple-level pretraining, can be recovered using only label-missing records, and remains complementary to protein language model representations. These findings demonstrate that incompletely labeled PROTAC databases contain useful relational supervision and provide a practical route for learning context-aware degradation predictors from scarce experimental labels.

Problem

• PROTAC
• degradation prediction
• context-aware
• label scarcity
• counterfactual pretraining

Innovation

• counterfactual tuple pretraining
• context-aware prediction
• PROTAC degradation
• self-supervised learning
• relational supervision