BEIJING, May 22, 2024 /PRNewswire/ — Docking is crucial in the early stages of drug discovery as it enables the efficient screening of vast libraries of compounds, saving time and resources. In recent years, AI-based methods have emerged as promising alternatives for molecular docking, offering the potential for high accuracy without incurring prohibitive computational costs. Advancing AI for Science, DP Technology has recently open-sourced its powerful AI-based Uni-Mol Docking v2 model [1] and made it available to the science community. Uni-Mol Docking v2 is a powerful docking model released prior to AlphaFold3 [2].
AI assisted Drug Discovery
Docking is a computational technique used in drug discovery to predict how "well" a small molecule (drug candidate) interacts with a target protein. Docking is crucial in the early stages of drug discovery as it enables the efficient screening of vast libraries of compounds, saving time and resources. Globally, the molecular docking market is a significant segment within the broader drug discovery informatics market, which is projected to grow from USD 3 billion in 2023 to USD 8 billion by 2032, driven by increasing investments in R&D and the adoption of advanced computational tools by pharmaceutical companies [3].
Docking techniques have evolved significantly from traditional physics/scoring-based methods to advanced deep learning / AI-based approaches. Initially, docking relied on physical and chemical principles to predict interactions, using scoring functions to evaluate binding affinities based on geometric and energetic considerations. While effective, these methods were computationally intensive and limited by their reliance on predefined rules. The advent of deep learning has transformed docking by enabling models to learn complex molecular representations directly from data. These deep learning models can capture intricate patterns and interactions that were previously hard to model, leading to more accurate and efficient predictions.
Uni-Mol Docking v2, is based on pre-trained AI model developed by DP Technology. The Uni-Mol modeling series, published at ICLR 2023 [4], described the pretraining of general molecular encoders and showcased their applications in various 2D and 3D downstream tasks such as molecular conformation generation, molecular property prediction and molecular docking. Uni-Mol achieved state-of-the-art results on a wide range of tasks, indicating its competence for generalization, and making it a powerful foundational model for molecular tasks.
In molecular docking, leveraging a pretrained molecular encoder, pretrained pocket encoder and joint pocket-ligand blocks, Uni-Mol Docking v2 achieved superior performance when compared to traditional docking algorithms like Autodock Vina in the CASF-2016 benchmark. Uni-Mol Docking v2 boasts an improved accuracy in predicting these binding poses with over 77% of ligands achieving an RMSD value under 2.0 Å and over 75% passing all quality checks. This marks a substantial leap from the 62% accuracy of our previous version, also eclipsing other known open-sourced methods. We’ve effectively tackled common challenges like chirality inversions and steric clashes, ensuring our predictions are not just accurate but also chemically viable.
Recently, the introduction of AlphaFold3 has been widely discussed in the scientific community. In this latest development, AlphaFold3 extends its capabilities to predict protein-ligand docking poses. In the AlphaFold3 paper on Nature, Uni-Mol Docking v2 is featured as a benchmark [2].
Committing to open science, DP Technology is proud to open-source Uni-Mol Docking v2 model, code, and dataset, making them available to the science community. We will continue to work on future iterations of Uni-Mol Docking and beyond as we commit to contributing to the global scientific community.
DP Technology is dedicated to advancing the frontiers of science through our unwavering commitment to open science. By openly sharing the research, data, and tools, DP aims to empower the global scientific community, accelerate discovery, and drive meaningful progress across various fields. DP’s commitment to transparency and collaboration is rooted in the conviction that the best solutions to the world’s challenges emerge when we work together, breaking down barriers to knowledge and fostering a culture of collective advancement.
Guided by this value, DP has been contributing actively and significantly to many open science projects in material science, drug discovery etc. They can be found at https://deepmodeling.com/. Recently, DP Technology is proud to join DeepModeling community in launching OpenLAM, towards developing the first Large Atom Model https://www.aissquare.com/openlam. DP has collaborated with global institutions to release DPA-2, which can accurately represent a diverse range of chemical systems and materials, enabling high-quality simulations and predictions with significantly reduced efforts compared to traditional methods[5].
Access Uni-Mol Docking v2
Ready-to-use Webapp: https://bohrium.dp.tech/apps/unimoldockingv2
Github: https://github.com/dptech-corp/Uni-Mol/tree/main/unimol_docking_v2
Paper: https://arxiv.org/abs/2405.11769
About DP Technology
DP Technology is a global leader in the "AI for Science" research paradigm, where AI learns scientific principles and data, then tackles key challenges in scientific research and industrial R&D.
DP’s commitment to interdisciplinary research has led to the creation of the DP "Particle Universe", an array of pre-trained large science models designed to bridge foundational research with practical industrial applications. DP’s software suite includes the Bohrium® Scientific Research Space, Hermite® Computational Drug Design Platform, RiDYMO® Dynamics Platform, and Piloteye® Battery Design Automation Platform. Together, these platforms form a robust foundation for industrial innovation and an open ecosystem for AI in science, fostering advancements in key areas such as drug discovery, energy, material science, and information technology.
Visit DP Technology at www.dp.tech/en
Reference
[1] https://arxiv.org/abs/2405.11769
[2] https://www.nature.com/articles/s41586-024-07487-w
[3] https://www.grandviewresearch.com/industry-analysis/drug-discovery-informatics-market ; https://www.alliedmarketresearch.com/drug-discovery-informatics-market-A07074;
[4] https://openreview.net/forum?id=6K2RM6wVqKu
[5] https://arxiv.org/abs/2312.15492
This content was prepared by our news partner, Cision PR Newswire. The opinions and the content published on this page are the author’s own and do not necessarily reflect the views of Siam News Network
