Qubit Pharmaceuticals has unveiled FeNNix-Bio1, a quantum AI foundation model developed to transform molecular simulations across biomolecular and pharmaceutical research. Trained entirely on synthetic quantum chemistry data, FeNNix-Bio1 integrates high-accuracy quantum methods including Density Functional Theory (DFT), Quantum Monte Carlo (QMC), and Configuration Interaction (CI) to build a comprehensive and generalizable representation of interatomic forces. By using transfer learning to combine the coverage of DFT with the precision of QMC and CI, the model captures quantum-level behavior in a scalable format. The approach was enabled by exascale high-performance computing resources provided by GENCI, EuroHPC, and Argonne.
FeNNix-Bio1 enables reactive molecular dynamics simulations at a scale and accuracy previously unattainable, including the formation and breaking of chemical bonds, proton transfer, and quantum nuclear effects. It has been validated on complex tasks such as simulating hydration free energies and predicting the physical behavior of water and solvated ions—critical for modeling biological systems. The model is capable of simulating million-atom systems over nanosecond timescales while maintaining quantum-level accuracy, which is not feasible with traditional force fields. Its capabilities allow researchers to study dynamic protein-drug interactions with high fidelity and extend beyond static structure prediction tools like AlphaFold by capturing the evolving nature of biomolecules.
The development of FeNNix-Bio1 reflects a cultural shift in computational science, uniting experts across quantum chemistry, machine learning, HPC, and molecular modeling. The team co-developed the software stack—including quantum solvers, neural training infrastructure, and simulation engines—to ensure vertical integration and performance optimization. This modular foundation model is adaptable across domains, including drug discovery, catalysis, and material science, and sets a precedent for leveraging quantum-accurate datasets to improve AI-driven simulation. Its scalability and adaptability also make it suitable for future refinement as better quantum data becomes available.
Access the official Qubit Pharmaceuticals blog announcement here and the full academic paper on arXiv here. A separate summary of the public release is available here.
May 20, 2025
