Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Pre-Training on Large-Scale Generated Docking Conformations with HelixDock to Unlock the Potential of Protein-ligand Structure Prediction Models

Lihang Liu, Shanzhuo Zhang, Donglong He, Xianbin Ye, Jingbo Zhou, Xiaonan Zhang, Yaoyao Jiang, Weiming Diao, Hang Yin, Hua Chai, Fan Wang, Jingzhou He, Liang Zheng, Yonghui Li, Xiaomin Fang

TL;DR

HelixDock tackles data scarcity in protein-ligand structure prediction by pre-training a geometry-aware model on a vast corpus of docking conformations generated by physics-based tools, then fine-tuning on a limited set of experimentally validated complexes. The two-phase architecture combines an InteractionLearner with a diffusion-based StructurePredictor to iteratively refine ligand coordinates within protein pockets, yielding physically plausible poses. Empirical results show HelixDock surpasses physics-based and prior deep-learning baselines on core benchmarks (PDBbind Core) and novel complexes (PoseBusters), with strong transferability across targets and protein families. The study reveals scaling laws where larger pre-training data and bigger models yield consistent improvements, and demonstrates practical benefits in cross-docking and structure-based virtual screening, underscoring HelixDock’s potential to accelerate drug discovery.

Abstract

Protein-ligand structure prediction is an essential task in drug discovery, predicting the binding interactions between small molecules (ligands) and target proteins (receptors). Recent advances have incorporated deep learning techniques to improve the accuracy of protein-ligand structure prediction. Nevertheless, the experimental validation of docking conformations remains costly, it raises concerns regarding the generalizability of these deep learning-based methods due to the limited training data. In this work, we show that by pre-training on a large-scale docking conformation generated by traditional physics-based docking tools and then fine-tuning with a limited set of experimentally validated receptor-ligand complexes, we can obtain a protein-ligand structure prediction model with outstanding performance. Specifically, this process involved the generation of 100 million docking conformations for protein-ligand pairings, an endeavor consuming roughly 1 million CPU core days. The proposed model, HelixDock, aims to acquire the physical knowledge encapsulated by the physics-based docking tools during the pre-training phase. HelixDock has been rigorously benchmarked against both physics-based and deep learning-based baselines, demonstrating its exceptional precision and robust transferability in predicting binding confirmation. In addition, our investigation reveals the scaling laws governing pre-trained protein-ligand structure prediction models, indicating a consistent enhancement in performance with increases in model parameters and the volume of pre-training data. Moreover, we applied HelixDock to several drug discovery-related tasks to validate its practical utility. HelixDock demonstrates outstanding capabilities on both cross-docking and structure-based virtual screening benchmarks.

Pre-Training on Large-Scale Generated Docking Conformations with HelixDock to Unlock the Potential of Protein-ligand Structure Prediction Models

TL;DR

HelixDock tackles data scarcity in protein-ligand structure prediction by pre-training a geometry-aware model on a vast corpus of docking conformations generated by physics-based tools, then fine-tuning on a limited set of experimentally validated complexes. The two-phase architecture combines an InteractionLearner with a diffusion-based StructurePredictor to iteratively refine ligand coordinates within protein pockets, yielding physically plausible poses. Empirical results show HelixDock surpasses physics-based and prior deep-learning baselines on core benchmarks (PDBbind Core) and novel complexes (PoseBusters), with strong transferability across targets and protein families. The study reveals scaling laws where larger pre-training data and bigger models yield consistent improvements, and demonstrates practical benefits in cross-docking and structure-based virtual screening, underscoring HelixDock’s potential to accelerate drug discovery.

Abstract

Protein-ligand structure prediction is an essential task in drug discovery, predicting the binding interactions between small molecules (ligands) and target proteins (receptors). Recent advances have incorporated deep learning techniques to improve the accuracy of protein-ligand structure prediction. Nevertheless, the experimental validation of docking conformations remains costly, it raises concerns regarding the generalizability of these deep learning-based methods due to the limited training data. In this work, we show that by pre-training on a large-scale docking conformation generated by traditional physics-based docking tools and then fine-tuning with a limited set of experimentally validated receptor-ligand complexes, we can obtain a protein-ligand structure prediction model with outstanding performance. Specifically, this process involved the generation of 100 million docking conformations for protein-ligand pairings, an endeavor consuming roughly 1 million CPU core days. The proposed model, HelixDock, aims to acquire the physical knowledge encapsulated by the physics-based docking tools during the pre-training phase. HelixDock has been rigorously benchmarked against both physics-based and deep learning-based baselines, demonstrating its exceptional precision and robust transferability in predicting binding confirmation. In addition, our investigation reveals the scaling laws governing pre-trained protein-ligand structure prediction models, indicating a consistent enhancement in performance with increases in model parameters and the volume of pre-training data. Moreover, we applied HelixDock to several drug discovery-related tasks to validate its practical utility. HelixDock demonstrates outstanding capabilities on both cross-docking and structure-based virtual screening benchmarks.
Paper Structure (16 sections, 2 equations, 6 figures, 1 table)

This paper contains 16 sections, 2 equations, 6 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methodology
  3. Generation of Large-scale Docking Complexes
  4. Training Paradism and Network Architecture of HelixDock
  5. Pre-training on Large-scale Docking Complexes Enhances Complex Structure Prediction
  6. Superior Performance of HelixDock Compared to Baseline Methods
  7. HelixDock demonstrates Its Transferability on Novel Complexes
  8. HelixDock Produces Physically Plausible Molecular Structures
  9. HelixDock Maintains High Accurate Predictions Across Various Protein Families
  10. Scaling Law for Protein-Ligand Structure Prediction
  11. Relations between the Performance and the Model Sizes
  12. Relations between the Performance and the Pre-training Data Sizes
  13. Practicality in Drug Discovery
  14. HelixDock Generalizes Well to Cross-Docking
  15. HelixDock Demonstrates Application Potential in Structure-Based Virtual Screening
  16. ...and 1 more sections

Figures (6)

  • Figure 1: Overall framework of HelixDock: deep learning-based protein-ligand structure prediction model enhanced by massive and diverse binding poses.
  • Figure 2: Comparison of HelixDock-Database and PDBbind.(a) Protein family comparison of HelixDock-Database and PDBbind. (b) UMAP of Morgan fingerprints of ligands from HelixDock-Database and PDBbind.
  • Figure 3: Overall evaluation of HelixDock and the baseline methods for complex structure prediction on complex prediction datasets. (a) Percentage of RMSD $\leq 2$Å (success rate) on PDBbind Core set. (b) Percentage of RMSD $\leq 1$Å on PDBbind Core set. (c) Percentage of RMSD $\leq 2$Å on PoseBusters benchmark. (d) Success rate in PoseBusters benchmark with cases stratified by sequence identity to the PDBbind 2020 General set. (e) Two examples from PoseBusters, PDB_id:7AA0 with a high sequence identity of 1.0 and PDB_id:6XM9 with a low sequence identity of 0.24. (f) PoseBuesters quality checks on predictions of DiffDock, AlphaFold-latest, and HelixDock. (g) Performance comparison across various protein families.
  • Figure 4: Scaling laws for protein-ligand structure prediction. (a) Scaling laws of model sizes in the pre-training stage. (b) Scaling laws of model sizes in the fine-tuning stage with models evaluated on PDBbind core set and PoseBusters benchmark. (c) Scaling laws of pre-training data sizes in the pre-training stage. (d) Scaling laws of pre-training data in the fine-tuning stage with models evaluated on PDBbind core set and PoseBusters benchmark.
  • Figure 5: Cross-docking. (a) Performance of HelixDock and other baselines in two cross-docking datasets, i.e., PDBbind-CrossDocked-Core and APObind-Core. (b) RMSD comparison of HelixDock on holo structures and apo structures. (c) Prediction of HelixDock on sample PDB ID:2XNB from PDBbind Core set when the protein is holo or apo.
  • ...and 1 more figures