High Quality Force Field Models for Biased Ligand Design

High Quality Force Field Models for Biased Ligand Design Molecular simulation (MS) plays an essential role in biochemical and biophysical research. A key element to successful applications of MS is the quality of a practical molecular mechanics force field (MMFF). The central mission of the Wang lab is to develop a set of MMFF models to accurately model the structures, dynamics, energetics, and functions of biomolecular systems. We aim to reduce the root-mean-square error of free energy prediction to be < 0.6 kcal/mol. To achieve this mission, we will first develop the 3rd generation of GAFF (GAFF3). GAFF3 development will be based on ABCG2, a new physical charge model which has demonstrated superior performance in large scale solvation free energy calculations. New force field parameterization techniques, such as applying machine learning potentials to fast detect “bad” torsional parameters, will be extensively applied in GAFF3 development. Special versions of GAFF3 which utilize a double exponential (DE) functional form for van der Waals interactions and new versions of GAFF2 with extended chemical space coverage will be developed in parallel. We will then critically evaluate the performance of newly developed GAFF force fields in studying biomolecule-ligand interactions using both pathway-based and endpoint free energy methods. Advanced sampling techniques including alchemical enhanced sampling and orthogonal space tempering will be applied for this evaluation effort. We believe that GAFF3 will be able to approach the performance limit an additive model can achieve. The successful development of GAFF3 will provide us with an advanced framework for addressing three challenges in the current drug design: (1) to efficiently and effectively describe the heterogenous environment where ligand-receptor binding occurs; (2) to rationally design a ligand which biasedly elicits a certain cellular signal but does not affect other pathways. A biased ligand is an attractive drug candidate as it can minimize unwanted or adverse effects; (3) to expand the druggable chemical space for those very promising drug targets. Artificial intelligence (AI) techniques will be utilized to power our biased ligand design. We have developed and evaluated a novel GAFF3-based scoring function, ligand-receptor interaction profile (LRIP) scoring function (LRIP-SF) to tackle the first challenge. LRIP-SF has achieved superior performance in both drug lead identification and optimization. As an accurate and efficient scoring function, LRIP-SF can fill the big gap between two widely used virtual screening methods, docking and alchemical free energy filters. We believe that the function and signaling pathways elicited by a ligand are encoded in the LRIP, and machine learning algorithms can learn the key attributes of the LRIP and generate scoring functions, LRIP-SFs, to recognize similar ligands in a screening library. Moreover, LRIP-SFs can be implemented into a generative modeling framework, such as generative adversarial networks (GAN) or diffusion models, to biasedly de novo design chemical structures. Thus, the AI-powered algorithms and DRUG-GAN models / diffusion models will be able to tackle the second and third challenges and likely revolutionize future drug discovery. Project Number: 1R35GM163906-01 | Fiscal Year: 2026 | NIH Institute/Center: National Institute of General Medical Sciences (NIGMS) | Principal Investigator: Junmei Wang | Institution: UNIVERSITY OF PITTSBURGH AT PITTSBURGH, PITTSBURGH, PA | Award Amount: $108,500 | Activity Code: R35 | Study Section: Special Emphasis Panel[ZRG1 MCST-G (56)] View on NIH RePORTER: https://reporter.nih.gov/project-details/11331325