A 17-year-old Indian-American student from Morgantown, West Virginia, is earning recognition for research that could help make drug discovery faster and more efficient. Pavan Subramani, a senior at Morgantown High School, has been named a 2026 Davidson Fellow and awarded a $25,000 scholarship for developing an artificial intelligence-based method to accelerate molecular simulations, according to his profile on the official Davidson Fellows website.His research combines generative diffusion models with molecular dynamics simulations, allowing scientists to study how proteins move and how potential drugs interact with them. In testing, his method sampled important molecular configurations 68 times more efficiently than conventional molecular dynamics for a specific measure of sampling efficiency. A related method also reached equilibrium ligand-binding populations more than four times faster than an existing technique.For Subramani, the recognition is both an acknowledgement of his work and an opportunity to pursue further studies in science.
How Pavan Subramani is using AI to advance drug discovery
Subramani’s project, titled The Hybridization of Generative Diffusion Models with Molecular Dynamics Simulations: A Novel Method to Accelerate Drug Discovery, addresses a challenge researchers face when studying potential medicines.Molecular dynamics simulations help scientists understand how proteins behave and how drug molecules bind to them. However, conventional simulations can be computationally expensive and may take a long time to capture rare molecular events, including changes in protein structure and shifts in the way a drug binds to a target.Subramani became interested in the problem while working on another drug discovery project involving potential inhibitors of the Receptor for Advanced Glycation End Products, a protein associated with several disease processes. After designing candidate molecules, he wanted to use molecular dynamics simulations to investigate whether they would bind to the target protein. The computational demands proved too high for his laptop, prompting him to explore enhanced sampling techniques.That experience led him to develop a framework that uses generative AI to suggest molecular changes while retaining the statistical safeguards needed for reliable simulations.
His method achieved 68 times greater sampling efficiency
Subramani developed a technique called diffusion hybridized molecular dynamics, or dhMD. It uses a diffusion model to propose changes to a protein’s structure and then applies a Metropolis-Hastings acceptance criterion to preserve the correct equilibrium distribution.In tests involving alanine dipeptide, the method delivered a 68-fold improvement in effective sample size per GPU hour for the slow phi dihedral compared with classical molecular dynamics. Effective sample size is a measure of how efficiently a simulation generates statistically useful samples.He subsequently extended the approach to ligand-binding simulations by developing diffusion hybridized BLUES, or dhBLUES. This method combines diffusion-based proposals with the existing BLUES enhanced-sampling technique, which helps researchers explore different ways a ligand can bind to a protein.In testing involving T4 lysozyme and toluene, dhBLUES converged to equilibrium ligand-binding populations more than four times faster than BLUES.The distinction is important: rather than simply generating molecular structures with AI, Subramani designed a way to integrate AI-generated proposals into simulations while preserving the statistical validity of the results.
Overcoming the challenge of keeping simulations accurate
One of the most difficult aspects of the research involved ensuring that AI-generated molecular changes would not introduce bias into the simulations.Subramani found that allowing a model to propose molecular changes directly from the current state could compromise the results. He addressed this by having the model predict the parameters of an explicit proposal distribution, from which the actual molecular move could be sampled. This made it possible to calculate the forward and reverse proposal probabilities needed for the acceptance criterion.He also encountered problems when large structural changes created molecular overlaps and extremely high energies, causing proposed moves to be rejected frequently. This challenge helped motivate the development of dhBLUES, which uses Nonequilibrium Candidate Monte Carlo to move a ligand through smaller steps while allowing its surroundings to relax.These technical developments form the basis of his broader approach: using generative models to make molecular simulations more efficient without sacrificing their underlying statistical principles.
Why the research could matter for future medicines
More efficient molecular simulations could help researchers evaluate potential drug candidates in less time and study molecular interactions that conventional simulations may struggle to capture.Understanding protein conformations and ligand-binding modes can provide valuable information about how strongly a molecule binds to a target and how selectively it interacts with it. Faster sampling could allow scientists to investigate more candidates and explore molecular behaviour with fewer computational resources.The work could also make advanced simulation techniques more accessible to academic laboratories that do not have access to extensive computing infrastructure. However, the findings represent a methodological advance in computational research; further work would be needed to establish how the approach performs across a wider range of proteins, drugs and drug discovery applications.Subramani’s research demonstrates how AI can be used not only to generate molecular structures but also to improve the computational methods scientists use to study them.
From international science recognition to a $25,000 scholarship
The Davidson Fellows scholarship recognises students who have completed significant work in science, technology, engineering, mathematics, literature, music or another eligible field. Subramani’s research has also received a third-place Grand Award in Chemistry at the International Science and Engineering Fair, and he has participated in the Massachusetts Institute of Technology’s Research Science Institute as a mentee.He plans to major in chemistry or applied physics and continue pursuing research in computational chemistry and biophysics, with a focus on advancing human health through drug discovery.Speaking about the scholarship, Subramani said, “To me, being selected as a Davidson Fellow recognizes that the research I am doing is valuable and encourages me to continue scientific research. The scholarship also allows me to pursue the higher education I will need to do this.”Alongside his scientific work, Subramani founded The Substance Safety Project, a nonprofit focused on substance abuse advocacy and education.According to his Davidson Fellows profile, the organisation has raised more than $10,000, distributed 750 drug disposal packets and conducted educational workshops reaching more than 1,500 people.