Subnet 25

Protein folding,
Open sourced

Proving the power of Distributed Compute

Subnet 25 Diagram

SN25 is our dedicated protein-folding subnet

Protein folding is one of the hardest problems in science, requiring vast compute power. While recent breakthroughs such as AlphaFold3 are very effective, their centralised compute means that cost per query is still too high.

Harnessing distributed compute drives down cost per query

SN25 gives scientists the tools they need to fold proteins. Its competitive design incentivises miners to develop protein-folding machine learning models as efficiently as possible, driving down cost even as performance improves.

Protein-folding is just the start

By demonstrating the advantages of distributed compute for solving complex challenges, SN25 is pioneering cutting-edge scientific research on the blockchain.

How SN25 Incentivises
Protein Folding

[1]

Benchmarked to best performance

SN25 uses the industry standard GROMACS software to simulate the molecular dynamics of proteins. Our base machine runs CUDA-supported GPUs as standard, raising the standard of baseline performance.

[2]

Motivating miners to relentless improvement

Miners compete to provide protein configurations that coincide with the lowest energy (analogous to loss). This aligns with the desired outcome of biologically stable structures, while also being deterministic and transparent.

[3]

Reducing compute, delivering results

SN25 shows that Bittensor's model can reduce the cost of even the most computationally intensive tasks - and deliver at scale, too. Since launch in June 2024, SN25 has already completed over 400,000 protein-folding jobs.