Synthetic Data Generation and Foundation Machine Learning Models for Molecular Simulation: Converging HPC and Quantum Computing

by Jean-Philip Piquemal (Laboratoire de Chimie Théorique, Sorbonne Université & Qubit Pharmaceuticals)

Tuesday Oct 7, 2025, 11:00 AM → 1:00 PM America/Chicago

Room 1407 (Building 240)

Foundation machine learning models emerge as a new pivotal tool for the simulation of molecular systems providing alternatives to ab initio molecular dynamics and to empirical force-fields. However, as they are solely designed on the basis of synthetic data, their accuracy strongly relies on the accuracy (and size) of their associated quantum chemistry reference databases. In that context, I will start by presenting our newly introduced FeNNiX-Bio1 foundation model, a scalable approach designed for accurate condensed phase simulations. Then I will detail our strategy for the high-performance production of high accuracy ab initio data. It encompasses two pillars: High Performance Computing (HPC) and Quantum Computing (QC). Therefore, I will start by introducing our GPU-accelerated implementations of the Diffusion Quantum Monte Carlo and selected-CI approaches. As they enable the fast evaluation of both energies and forces, I will put them in the context of the required large-scale production of high accuracy synthetic data. Then I will introduce our efforts towards quantum computing (QC). Indeed, QC holds the potential to outperform classical computing for simulating quantum many-body systems. However, the practical implementation of its key algorithms on current quantum processing units faces challenges in measuring a polynomially scaling number of observables during the operator selection so as to optimise a high-dimensional and noisy cost function. In this talk, I will present various quantum algorithms enabling to perform accurate quantum chemistry computations. More precisely, I will present their implementation on the Hyperion-1 scalable quantum emulator platform. Indeed, it enables to test such algorithms at scale as it offers the access to large number of logical qubits using present GPU-accelerated HPC infrastructures.

[1] A Foundation Model for Accurate Atomistic Simulations in Drug Design.T. Plé, O. Adjoua, A. Benali, E. Posenitskiy, C. Villot, L. Lagardère, J.-P.Piquemal, 2025, submitted, DOI: 10.26434/chemrxiv-2025-f1hgn-v3

[2] Pushing the Accuracy Limit of Foundation Neural Network Models with Quantum Monte Carlo Forces and Path Integrals .A. Benali, T. Plé, O. Adjoua,  V. Agarawal, T. Applencourt, M. Blazhynska, R. Clay III, K. Gasperich, K. Hossain, J. Kim, C. Knight, J. T. Krogel, Y. Maday, M. Maria, M. Montes, Y. Luo, E. Posenitskiy, C. Villot, V. Vishwanath, L. Lagardère,  J.-P. Piquemal, 2025, DOI: 10.48550/arXiv.2504.07948

       [3] Overlap-ADAPT-VQE: Practical Quantum Chemistry on Quantum Computers via Overlap-Guided Compact Ansätze. C. Feniou, M. Hassan, D. Traoré, E. Giner, Y. Maday, J.-P. Piquemal, Commun. Phys., 2023, 6, 192 (Open Access), DOI: 10.1038/s42005-023-01312-y

       [4] Shortcut to Chemically Accurate Quantum Computing via Density-based Basis-set Correction. D. Traore, O. Adjoua, C. Feniou, I.-M. Lygatsika, Y. Maday, E. Posenitskiy, K. Hammernik,  A. Peruzzo, J. Toulouse, E. Giner, J.-P. Piquemal,  Commun. Chem., 2024, 7, 269 (Open Access), DOI: 10.1038/s42004-024-01348-3

      

Slides