zPoseScore model for accurate and robust protein–ligand docking pose scoring in CASP15

Abstract

We introduce a deep learning‐based ligand pose scoring model called zPoseScore for predicting protein–ligand complexes in the 15th Critical Assessment of Protein Structure Prediction (CASP15). Our contributions are threefold: first, we generate six training and evaluation data sets by employing advanced data augmentation and sampling methods. Second, we redesign the “zFormer” module, inspired by AlphaFold2's Evoformer, to efficiently describe protein–ligand interactions. This module enables the extraction of protein–ligand paired features that lead to accurate predictions. Finally, we develop the zPoseScore framework with zFormer for scoring and ranking ligand poses, allowing for atomic‐level protein–ligand feature encoding and fusion to output refined ligand poses and ligand per‐atom deviations. Our results demonstrate excellent performance on various testing data sets, achieving Pearson's correlation = 0.783 and 0.659 for ranking docking decoys generated based on experimental and predicted protein structures of CASF‐2016 protein–ligand complexes. Additionally, we obtain an averaged local distance difference test (lDDT pli = 0.558) of AIchemy LIG2 in CASP15 for de novo protein–ligand complex structure predictions. Detailed analysis shows that accurate ligand binding site prediction and side‐chain orientation are crucial for achieving better prediction performance. Our proposed model is one of the most accurate protein–ligand pose prediction models and could serve as a valuable tool in small molecule drug discovery.

Recent Progress of Protein Tertiary Structure Prediction

“…The Zou group [ 142 ] adopted a similar strategy, integrating the physicochemical molecular docking method AutoDock Vina [ 143 ] with the ligand similarity methodology SHAFTS [ 144 ]. In the Alchemy_LIG team [ 145 ] protein structures were constructed using AlphaFold2, and ligands were docked utilizing the AutoDock Vina docking method and a machine learning model trained to detect native binding modes. The ClusPro group [ 146 ] employed AlphaFold2 for constructing monomer protein structures and created multimeric assemblies via a template-based docking algorithm, ClusPro LigTBM [ 146 ], for general ligand placement, alongside the Glide program [ 147 ], for direct docking in cases when no templates were found.…”

Fully Flexible Molecular Alignment Enables Accurate Ligand Structure Modeling

“…Traditional scoring functions estimate the binding affinity or calculate the binding free energies of the complexes using either knowledge-based parametrization, statistical modeling, − or force-field-based functions . In addition, a lot of the machine-learning-based or deep-learning-based scoring functions − have been developed in recent years. These models are generally used as a postscore method to rescore the poses of the ligand generated by various docking applications.…”

Fully flexible molecular alignment enables accurate ligand structure modelling

“…Traditional scoring functions estimate the binding affinity or calculate the binding free energies of the complexes using either knowledge-based parameterization, statistical modeling [7,8,9,10], or force fieldbased functions [11]. In addition, a lot of the machine learning-based or deep learning-based scoring functions [12,13,14,15,16,17,18,19] have been developed in recent years. These models are generally used as a post-scoring method to re-score the poses of the ligand generated by various docking applications.…”