Online triplet contrastive learning enables efficient cliff awareness in molecular activity prediction

Abstract

Predicting molecular activity and quantitative structure-activity relationship (QSAR) is important for drug discovery and optimization. With molecular structures as frames, graph neural networks (GNNs) are suited for activity prediction but tend to overlook activity-cliffs (ACs) where structurally-similar molecules have vastly different activity values. To address this, we introduced a new online triplet contrastive learning framework ACANet that incorporates a unique activity-cliff-awareness (ACA) loss function, enabling efficient AC-awareness during training. The ACA loss enhances metric learning in the latent space and task learning in the target space simultaneously to make networks aware of ACs. ACANet outperformed the state-of-the-art machine learning and deep learning models in activity prediction and AC awareness on 39 benchmark datasets. ACA loss function is superior in AC-awareness than the mean absolute error and mean squared error loss functions. This innovative approach opens new avenues and provides valuable tools for applications in drug discovery and chemical engineering.

Enhancing Generalizability in Protein–Ligand Binding Affinity Prediction with Multimodal Contrastive Learning

“…For a biased model, it may correctly predict the binding affinity based on the incorrect model. Noteworthy, contrastive learning has showcased competitive results in tasks like small molecule property prediction, − sequences-based prediction of drug–target interactions, similarity-based virtual screening, reaction classification, and enzyme function prediction . Contrastive learning can be applied in a multimodal setting, which enhances the learning of joint representations from varied modalities and thus bolsters model’s performance. , Given those considerations, we have therefore employed contrastive learning to reduce the embedding distance of both protein–ligand representation modalities and expand the distance from incorrect poses in a shared latent space, thereby improving the protein–ligand representation.…”

OLB-AC: toward optimizing ligand bioactivities through deep graph learning and activity cliffs

“… Winkler and Le (2017) investigated how deep neural networks can address challenges posed by ACs in QSAR datasets. Shen et al 2023 introduced a new online triplet contrastive learning framework (ACANet) to enable efficient AC awareness in molecular activity prediction. Yin et al (2022a ) proposed AFSE to improve generalization performance and reduce the impact of ACs on bioactivity prediction.…”