Benchmarking Simulacra AI's Quantum Accurate Synthetic Data Generation for Chemical Sciences

2025/10/30 by Fabio Falcioni, Е Е Орлова, Falcioni, Fabio +7
Materials Science · Decision Sciences · Biochemistry, Genetics and Molecular Biology · #Machine Learning in Materials Science #Scientific Computing and Data Management #Protein Structure and Dynamics

paper · pdf · doi:10.48550/arxiv.2511.07433

Abstract

In this work, we benchmark \simulacra's synthetic data generation pipeline against a state-of-the-art Microsoft pipeline on a dataset of small to large systems. By analyzing the energy quality, autocorrelation times, and effective sample size, our findings show that Simulacra's Large Wavefunction Models (LWM) pipeline, paired with state-of-the-art Variational Monte Carlo (VMC) sampling algorithms, reduces data generation costs by 15-50x, while maintaining parity in energy accuracy, and 2-3x compared to traditional CCSD methods on the scale of amino acids. This enables the creation of affordable, large-scale ab-initio datasets, accelerating AI-driven optimization and discovery in the pharmaceutical industry and beyond. Our improvements are based on a novel and proprietary sampling scheme called Replica Exchange with Langevin Adaptive eXploration (RELAX).

Citations

• How Accurate Are DFT Forces? Unexpectedly Large Uncertainties in Molecular Datasets
• The Open Catalyst 2025 (OC25) Dataset and Models for Solid-Liquid Interfaces
• An ab initio foundation model of wavefunctions that accurately describes chemical bond breaking
• The Open Molecules 2025 (OMol25) Dataset, Evaluations, and Models
• Accurate Ab-initio Neural-network Solutions to Large-Scale Electronic Structure Problems
• Beyond CCSD(T) accuracy at lower scaling with auxiliary field quantum Monte Carlo
• A Survey on Universal Approximation Theorems
• Neural Pfaffians: Solving Many Many-Electron Schrödinger Equations
• Transferable Neural Wavefunctions for Solids
• A foundation model for atomistic materials chemistry
• Neural network variational Monte Carlo for positronic chemistry
• Accurate computation of quantum excited states with neural networks
• Variational Monte Carlo on a Budget -- Fine-tuning pre-trained Neural Wavefunctions
• Generalizing Neural Wave Functions
• A Self-Attention Ansatz for Ab-initio Quantum Chemistry
• DiffDock: Diffusion Steps, Twists, and Turns for Molecular Docking
• SPICE, A Dataset of Drug-like Molecules and Peptides for Training Machine Learning Potentials
• MACE: Higher Order Equivariant Message Passing Neural Networks for Fast and Accurate Force Fields
• Local-Global MCMC kernels: the best of both worlds
• Ab-Initio Potential Energy Surfaces by Pairing GNNs with Neural Wave Functions
• Spectral Gap of Replica Exchange Langevin Diffusion on Mixture Distributions
• Thirty years of density functional theory in computational chemistry: an overview and extensive assessment of 200 density functionals
• Non-convex learning via Stochastic Gradient Langevin Dynamics: a\n nonasymptotic analysis
• The Python-based Simulations of Chemistry Framework (PySCF)
• An efficient variational principle for the direct optimization of excited states
• Zur Quantentheorie der Molekeln

Related

• Re-evaluating sample efficiency in de novo molecule generation
• Augmented and Programmatically Optimized LLM Prompts Reduce Chemical Hallucinations
• The Convergence Frontier: Integrating Machine Learning and High Performance Quantum Computing for Next-Generation Drug Discovery
• Hybrid Quantum Generative Adversarial Networks for Molecular Simulation and Drug Discovery
• Amortized Sampling with Transferable Normalizing Flows