May 1, 2025
Neural network potentials (NNPs) are revolutionizing molecular simulation. Starting today, Rowan is teaming up with Macrocosmos, an open-source AI research lab building on Bittensor, to accelerate the development of the next generation of NNPs through Subnet 25 - Mainframe.
From Macrocosmos' documentation on Subnet 25 - Mainframe:
Mainframe is a decentralised science subnet on Bittensor. It provides computing power and community talent to solve scientific problems.
Subnet 25 currently tackles decentalized protein folding using molecular dynamics (MD) a method for simulating the physical movements of atoms and molecules.
Our collaboration brings Bittensor closer to cutting-edge research in drug discovery and materials science. We look forward to working with the Macrocosmos team to add new capabilities to Mainframe.
In our newly released preprint "Egret-1: Pretrained Neural Network Potentials For Efficient and Accurate Bioorganic Simulation," we present a new family of open-source ML models for high-accuracy bioorganic simulations. One key insight from the paper is that the next generation of models will need a lot more high-quality data generated via density-functional theory (DFT). From the paper:
In particular, we anticipate that a combination of improved dataset scale and quality, more expressive architectures, and performance optimization will make it possible to achieve significantly improved accuracy, speed, and generality, which we expect to have a substantial impact on discovery across the chemical sciences.
Working with Macrocosmos and Mainframe will let us use the decentralized computing power of the Bittensor network to dynamically generate the data we need to train the next generation of NNPs.
Bittensor is an ideal partner for this ambitious scientific project. We're very excited to be working with Macrocosmos, Subnet 25 - Mainframe, and the community to combine our expertise and push the boundaries of ML-powered molecular simulation forward.