Uni-Mol Docking V2: Towards Realistic and Accurate Binding Pose Prediction

TL;DR

This work tackles the gap between quantitative docking metrics and chemical-physical plausibility by introducing Uni-Mol Docking V2. Leveraging a reproducible data pipeline, pretrained molecular and pocket encoders, and a hybrid framework with physics-based Uni-Dock, the approach achieves state-of-the-art PoseBusters performance with RMSD $< 2.0\,\mathrm{Å}$ for $77+\%$ of ligands and $75+\%$ quality-passing cases, while delivering $>95\%$ chemically plausible predictions and reducing unphysical artifacts. The results demonstrate improved chemical accuracy and robust high-quality predictions, especially when integrated with physics-based refinements, across diverse targets. This positions Uni-Mol Docking V2 as a powerful, industry-relevant tool for virtual screening and rational drug design, with open code, data, and services available to the community.

Abstract

In recent years, machine learning (ML) methods have emerged as promising alternatives for molecular docking, offering the potential for high accuracy without incurring prohibitive computational costs. However, recent studies have indicated that these ML models may overfit to quantitative metrics while neglecting the physical constraints inherent in the problem. In this work, we present Uni-Mol Docking V2, which demonstrates a remarkable improvement in performance, accurately predicting the binding poses of 77+% of ligands in the PoseBusters benchmark with an RMSD value of less than 2.0 Å, and 75+% passing all quality checks. This represents a significant increase from the 62% achieved by the previous Uni-Mol Docking model. Notably, our Uni-Mol Docking approach generates chemically accurate predictions, circumventing issues such as chirality inversions and steric clashes that have plagued previous ML models. Furthermore, we observe enhanced performance in terms of high-quality predictions (RMSD values of less than 1.0 Å and 1.5 Å) and physical soundness when Uni-Mol Docking is combined with more physics-based methods like Uni-Dock. Our results represent a significant advancement in the application of artificial intelligence for scientific research, adopting a holistic approach to ligand docking that is well-suited for industrial applications in virtual screening and drug design. The code, data and service for Uni-Mol Docking are publicly available for use and further development in https://github.com/dptech-corp/Uni-Mol.

Figures (5)

• Figure 1: Framework of Uni-Mol Docking V2
• Figure 2: Performance of different ML docking methods on the PoseBusters test set
• Figure 3: Comparison of plausibility checks for predictions by DiffDock, AlphaFold latest, and Uni-Mol Docking V2
• Figure 4: Waterfall plot showcasing cumulative impact of errors
• Figure 5: a: 7PRM, RMSD=1.11 Å. Inhibition of Monoacylglycerol lipase (MAGL), a key enzyme in the endocannabinoid system, has been proposed as an attractive approach for the treatment of various diseases including neurodegeneration, psychiatric disorders, and cancer; b: 8C7Y, RMSD=0.38 Å. BRAF inhibitors have revolutionized treatment of some cancers such as melanoma, although some undesired effects have been seen, such as the paradoxical hyperactivation of MAPK caused by the ligand-induced dimerization of 1st gen BRAF inhibitors. c: 7R9N. RMSD=0.63 Å. Hematopoietic progenitor kinase 1 (HPK1) is implicated as a negative regulator of T-cell receptor-induced T-cell activation. Inhibition of HPK1 has been shown to increase T-cell antitumor response. d: 7XI7, RMSD=1.25 Å. Novel inhibitors for human dihydrofolate reductase could expand therapeutic options against the parasitic toxoplasmosis infections. e: 7NP6, RMSD=0.54 Å. Inhibition of the nuclear receptor retinoic-acid-receptor-related orphan receptor $\gamma \text{t}$ (ROR$\gamma \text{t}$) is a promising strategy in the treatment of autoimmune diseases. f: 7N03, RMSD=1.03 Å. MTH1 is a DNA damage control enzyme and potentially synthetic lethal target. Its inhibition could open new avenues in oncologic targeted therapy.