Courtesy Photo | DIFFDOCK employs a diffusion process to iteratively refine a ligand's position within...... read more read more
Subscribe
An innovative deep learning model called DIFFDOCK offers a promising new approach to accelerate drug discovery and develop medical countermeasures (MCMs) to protect the Joint Force against new and emerging biological threats.
DIFFDOCK is a new molecular docking approach model for drug discovery, which has led to an even more advanced method—DIFFDOCK-L—that uses increased data-generating techniques and a larger model size with higher accuracy in predicting ligand binding poses. Overall, DIFFDOCK and DIFFDOCK-L represent a significant step forward in molecular docking, offering better accuracy, efficiency, and flexibility compared to traditional search-based molecular docking methods.
The Defense Threat Reduction Agency’s (DTRA) Chemical and Biological Technologies Department in its role as the Joint Science and Technology Office (JSTO) for Chemical and Biological Defense, an integral component of the Chemical and Biological Defense Program, is investing in basic research at the Massachusetts Institute of Technology (MIT) that integrates computer-aided drug development (CADD) techniques, such as computational modeling, complex algorithms, computer software, and molecular docking approaches, to enhance the rapid development of effective MCMs.
CADD approaches bypass the slow, traditional drug discovery and development process by shortening the timeline to improve existing or develop new MCMs against biological pathogens. Molecular docking predicts how well a drug molecule (ligand) will fit and bind to its target protein (e.g., protein of a pathogen). This is crucial in drug discovery as it shows key interactions between a drug and its target protein, revealing the molecular basis of its activity.
Traditional molecular docking methods predict the optimal binding pose (orientation) of a ligand and estimate its binding affinity with the protein These methods involve searching through many possible ligands poses until the best one is found. However, search-based methods can be expensive and difficult to scale to large datasets, plus the accuracy can be limited by the scoring function and search algorithm.
Through DTRA JSTO’s investments in basic research, researchers at MIT’s Computer Science & Artificial Intelligence Laboratory generated this new approach that differs from previous regression-based frameworks to use a generative modeling approach that is better aligned with the objective of molecular docking. At the heart of its modelling strategy, DIFFDOCK and DIFFDOCK-L learn the entire range of possible poses that a ligand might adopt when bound to a protein rather than predict a single best one through a process that refines random poses according to their compatibility with the protein structure.
By using a learned score, DIFFDOCK guides the refinement process to ensure compatibility with the protein. Over time, the random initial pose of the ligand transforms into a pose that better fits the protein. This diffusion process allows DIFFDOCK and the improved and advanced DIFFDOCK-L to explore a wider range of possibilities and ultimately achieve high accuracy in predicting ligand poses and allow them to generate new poses that a ligand might adopt when bound to the protein of a pathogen. The ability to learn from a distribution of the ligand’s poses and the ability to guide the refinement process using a learned score allows DIFFDOCK and DIFFDOCK-L to explore a wide range of possibilities and achieve high accuracy in predicting ligand poses and providing additional options for new drugs against biological threats.
Although the choice between traditional search-based molecular docking approaches and DIFFDOCK or DIFFDOCK-L depends on the specific research question, computational resources, and available data, DIFFDOCK and DIFFDOCK-L offer a promising new approach that can accelerate drug discovery and help to develop new treatments to protect the Joint Force against biological pathogens.
Sidebar-1
Molecular docking is a computational method in drug discovery to predict the preferred orientation, conformation, and binding affinity of two interacting molecules, typically ligand (a drug candidate) and target (a biological protein). In addition, molecular docking can guide optimizing the ligand’s structure to improve its binding properties to the protein. Molecular docking plays a crucial role in identifying potential drug candidates and is often used with experimental techniques to lead the design and optimization of new drugs.
Sidebar-2
Unlike traditional molecular docking methods, which often use a search-based and scoring function approach, DIFFDOCK is a new approach to molecular docking that uses diffusion generative modeling, aimed at to learn the entire range of possible poses (positions, orientations, and conformations) that a ligand might adopt when bound to a target protein. DIFFDOCK provides faster and more precise computer-aided design of small molecules and will contribute to the discovery of new promising candidates against biological threats.
POC: Annette von dem Bussche-Huennefeld, PhD, annette.e.vondembussche-huennefeld.civ@mail.mil
| Date Taken: | 08.12.2024 |
| Date Posted: | 08.12.2024 22:58 |
| Story ID: | 478444 |
| Location: | FT. BELVOIR, VIRGINIA, US |
| Web Views: | 187 |
| Downloads: | 0 |
This work, On Demand Solutions: How Innovations in Artificial Intelligence and Data Science are Delivering Biological Treatments, must comply with the restrictions shown on https://www.dvidshub.net/about/copyright.
No keywords found.
