Active Site Sequence Representations of Human Kinases Outperform Full Sequence Representations for Affinity Prediction and Inhibitor Generation: 3D Effects in a 1D Model

Abstract

Recent advances in deep learning have enabled the development of large-scale multimodal models for virtual screening and de novo molecular design. The human kinome with its abundant sequence and inhibitor data presents an attractive opportunity to develop proteochemometric models that exploit the size and internal diversity of this family of targets. Here, we challenge a standard practice in sequence-based affinity prediction models: instead of leveraging the full primary structure of proteins, each target is represented by a sequence of 29 discontiguous residues defining the ATP binding site. In kinase–ligand binding affinity prediction, our results show that the reduced active site sequence representation is not only computationally more efficient but consistently yields significantly higher performance than the full primary structure. This trend persists across different models, data sets, and performance metrics and holds true when predicting pIC50 for both unseen ligands and kinases. Our interpretability analysis reveals a potential explanation for the superiority of the active site models: whereas only mild statistical effects about the extraction of three-dimensional (3D) interaction sites take place in the full sequence models, the active site models are equipped with an implicit but strong inductive bias about the 3D structure stemming from the discontiguity of the active sites. Moreover, in direct comparisons, our models perform similarly or better than previous state-of-the-art approaches in affinity prediction. We then investigate a de novo molecular design task and find that the active site provides benefits in the computational efficiency, but otherwise, both kinase representations yield similar optimized affinities (for both SMILES- and SELFIES-based molecular generators). Our work challenges the assumption that the full primary structure is indispensable for modeling human kinases.

On the Choice of Active Site Sequences for Kinase-Ligand Affinity Prediction

“…The experimental setup is largely identical to the binding affinity prediction task described in ref ( 7 ). We take data from BindingDB 11 and examine two types of models, a k -nearest-neighbor (KNN) model that builds a joint similarity space of protein and ligand distances and a deep neural network called BiMCA (Bimodal Multiscale Convolutional Attention encoder 12 ) that ingests protein and ligand sequences (SMILES strings) and consists of convolutional and attention layers.…”

“…In our previous work, 7 the active site representation relied on 29 residues defined originally in Sheridan et al [ref ( 8 ), Table 1]. These residues are short contiguous subsequences that lie discontiguously in the original sequence (cf.…”

“…The superiority of the active site representation manifested consistently across all ligand types, with the sole exception of one drug class: MEK/MAPK inhibitors. 7 Notably, this class contains many allosteric binders, in particular ATP-noncompetitive MAPK inhibitors that bind to a unique site near the ATP-binding pocket. 9 One goal of the presented work is to address this systematic limitation in modeling allosteric binders and refine the definition of an “active site” for binding affinity prediction.…”

XLPFE: A Simple and Effective Machine Learning Scoring Function for Protein–Ligand Scoring and Ranking

“…This is consistent with previous studies, which showed that the use of active site sequences can improve the prediction of the affinity. 68 , 69 For the occurrence of amino acids in the protein active site, it can be seen that His, Gly, Tyr, and Trp rank in the top position ( Figure S3 ), probably because they can form hydrogen bonds with ligands and thus play key roles in ligand binding. Leu and Phe also rank quite high, possibly due to their contribution to the formation of hydrophobic pockets.…”

Improving the Assessment of Deep Learning Models in the Context of Drug-Target Interaction Prediction

“…All the datasets in this work are derived from BindingDB (Liu et al, 2007): a publicly accessible and regularly updated collection of binding affinity values between proteins considered to be drugtargets, and drug-like molecules. In particular, we adopt two benchmark datasets derived from BindingDB, one released by Yingkai Gao et al (2018) and the other as defined by Karimi et al (2019), which have been used for benchmarking recent DTI predictors (Chen et al, 2020; Born et al, 2022). Both benchmark datasets are outlined in Table 1.…”