Recursion and MIT unveil AI model to accelerate drug discovery

USA – Recursion, in collaboration with researchers at the Massachusetts Institute of Technology (MIT), has launched Boltz-2, a new open-source AI model designed to enhance drug discovery.

This technology enables scientists to predict molecular structure and binding affinity, thereby accelerating and improving the process.

Boltz-2 builds on Boltz-1, the first fully open-source AI model released last year, which competed with Google DeepMind’s AlphaFold 3 in structural predictions.

The latest version takes a step further by jointly modeling 3D biomolecular structures and assessing the strength of a drug’s attachment to its target, a crucial factor in determining drug effectiveness.

Drug developers often struggle to predict molecular binding strength, which is key to identifying promising treatments.

Regina Barzilay, an AI and health expert at MIT, highlighted that Boltz-2 not only improves accuracy but also opens new possibilities for exploring biological interactions that were previously too complex.

Powerful AI computing for faster results

Trained on Recursion’s NVIDIA-powered BioHive-2 supercomputer, Boltz-2 delivers predictions at a speed more than 1000 times faster than physics-based free energy perturbation (FEP) methods, which are commonly used to estimate molecular binding strength.

Recursion, which merged with Exscientia last year, emphasized that the decreased cost and increased speed of Boltz-2 makes large-scale virtual drug screening more feasible than ever before.

Boltz-2 is open-source, meaning that academic and commercial researchers can adapt it for specific types of molecules, making it an even more valuable tool for accelerating drug discovery.

“Because Boltz-2 includes its training code, scientists can refine the model for their needs, further advancing molecular research,” Barzilay added.

With its precision modeling, rapid results, and accessibility, Boltz-2 could transform how researchers develop new treatments, making drug discovery faster and more effective.