Contrastive learning in protein language space predicts interactions between drugs and protein targets

Abstract

Sequence-based prediction of drug–target interactions has the potential to accelerate drug discovery by complementing experimental screens. Such computational prediction needs to be generalizable and scalable while remaining sensitive to subtle variations in the inputs. However, current computational techniques fail to simultaneously meet these goals, often sacrificing performance of one to achieve the others. We develop a deep learning model, ConPLex, successfully leveraging the advances in pretrained protein language models (“PLex”) and employing a protein-anchored contrastive coembedding (“Con”) to outperform state-of-the-art approaches. ConPLex achieves high accuracy, broad adaptivity to unseen data, and specificity against decoy compounds. It makes predictions of binding based on the distance between learned representations, enabling predictions at the scale of massive compound libraries and the human proteome. Experimental testing of 19 kinase-drug interaction predictions validated 12 interactions, including four with subnanomolar affinity, plus a strongly binding EPHB1 inhibitor ( K D = 1.3 nM). Furthermore, ConPLex embeddings are interpretable, which enables us to visualize the drug–target embedding space and use embeddings to characterize the function of human cell-surface proteins. We anticipate that ConPLex will facilitate efficient drug discovery by making highly sensitive in silico drug screening feasible at the genome scale. ConPLex is available open source at https://ConPLex.csail.mit.edu .

In Silico Structural Modeling and Binding Site Analysis of Cerebroside Sulfotransferase (CST): A Therapeutic Target for Developing Substrate Reduction Therapy for Metachromatic Leukodystrophy

“…In the absence of experimentally determined protein 3D structures, computational modeling plays a cost-effective role in structure-based drug discovery. − In this study, for 3D model generation, a fully automated protein structure homology modeling server SWISS-MODEL was used. One of the advantages of SWISS-MODEL over other modeling software is that it follows the proper natural assembly of the template protein, expels all anticorrosive amino residues, and further adjusts the target-template arrangement in DeepView program. , One important feature of SWISS-MODEL is ProMod3, a central script-based platform for increasing the accuracy of the produced model as well as for quality estimation based on QMEANDisCo Global values.…”

Inference of drug off-target effects on cellular signaling using Interactome-Based Deep Learning

“…Similarly, another approach called ChemCPA 54 also encodes the chemical structure of the drug and non-linearly scales its dose and combines it with the drug representation. On this front, our drug module could also incorporate a non-linear encoder to represent the chemical structure of drugs and combine it with targets' representations, by building upon ideas presented in OmegaFold 55 and AlphaFold2 56 , in order to infer potential drug-target interactions, similar to what has been recently proposed in the ConPLex model 57 , after training models to ultimately predict the transcriptional profile of a cell.…”

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

“…We use a tanh decay with a restart schedule to gradually reduce this margin. 52 Specifically, for every E max = 10 contrastive training epochs, the margin is reset to its initial value. This allows us to dynamically adjust the margin throughout training for a better model performance (eq 16).…”

“…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, 49−51 sequences-based prediction of drug−target interactions, 52 similarity-based virtual screening, 53 reaction classification, 54 and enzyme function prediction. 55 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.…”

“…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.…”