Open-source, autonomous, & collaborative: understanding SN25’s approach

Looking towards subnet 25's future.

What is the best way to optimize decentralized protein folding? This is the question at the heart of SN25. We’ve been refining our competition and its output to ensure our subnet contributes to the life sciences space and DeSci community. While the process of folding proteins is intrinsically rewarding, if we want to make the greatest impact, then we must build something accessible, autonomous, and capable of being used by researchers.

We’ve been sculpting SN25 into a tool that can contribute to scientific research beyond the Bittensor community. To do that, we’ve needed to think differently about subnet design, and long-term plans.

Shifting to OpenMM

On September 24th, we moved SN25 over to OpenMM from GROMACS. This was for several reasons - some practical, and some ideological.

Strengthening our output

On the practical side, OpenMM provides our subnet with greater reproducibility. This means that operations are easier to replicate across different runs, environments, and hardware setups. Alongside this, it makes validation much more foolproof, meaning it functions better in a decentralized setting. This set us up well to engage in a greater diversity of proteins along with the systems that parameterize the simulations, giving SN25 more capacity to explore novel folding pathways and study a wider range of biomolecular behaviors.

Added to this is the fact that it makes it easier to gain hyper-parameter control on jobs and customizable force fields. This means that users can fine-tune simulations to match specific research needs, adjust variables like temperature and pressure, and mimic real-world conditions more simply.

Supporting our community

Another reason we chose OpenMM is because it is a Pythonic engine. It’s designed for Python users, has its own Python-based API, and can easily integrate with other Python libraries (such as NumPy, one we use regularly at SN25). Working with Python was important to us as it is a much more human-readable language, which is preferable compared to GROMACS’ primarily command-line workflow.

As a pioneer of decentralization, we wanted to make SN25 as accessible as possible. Therefore, we adopted an engine with a lower barrier to entry. Plus, Python is favored in the machine learning world, and while this subnet doesn’t use machine learning yet, it will in the future. Being within the Bittensor ecosystem also means there will be many people working with us who know Python already, so using OpenMM will have a smaller learning curve for them.

Building SN25 poses a very different challenge to those of many other subnets. We want to be serious contributors to the life sciences and DeSci world, but simultaneously we need to create something that our community can engage with. Therefore, OpenMM felt like the obvious choice, as it not only offers practical improvements, but also because its Pythonic nature makes it more accessible to people in this space.

Collaborating with Researchers

Our transition to OpenMM places us in the prime position to approach researchers. The improvements this engine added to our output revealed how powerful SN25 is, and so we’ve begun chatting with those in the life sciences space to learn how our subnet can help complement their work. We see forming these bonds as vital as it further connects us to STEM, and gives us the opportunity to help improve the world.

The goal for SN25 is for it to have a marked impact. We are building something that scientists can actively use in their work and studies. Seeing them engage with our subnet and think up use cases with us has been eye-opening.

Not only that, but the potential for cross-pollination between departments is huge. Giving researchers the room to share with us what innovations they’re working on, and for us to introduce them to the distributed, permissionless, and revolutionary happenings in Bittensor is rewarding. It places us at the forefront of Bittensor and DeSci.

A global job pool of organic queries

For our protein-folding subnet to have the greatest impact, it needs to be capable of taking organic queries – requests for tasks made by individuals. At present, SN25 operates by setting challenges to miners that come from synthetic data. This is typical for many subnets, as the process of finding fresh real-world data to form challenges would be taxing and slow.

However, we are transforming SN25 so that organic queries can be made, and miners can choose which ones they wish to work on. Synthetic queries will still be a major part of SN25, but in due course, organic queries will be available, too. This allows universities, research institutes, and companies to come to our subnet and request specific types of proteins be folded, with miners getting rewarded by these organizations.

Miners will be able to see which jobs are currently available and can pick ones that suit their hardware and optimized setup, giving them greater autonomy over how their computing power is used. Plus, they can potentially earn more emissions depending on tasks, as some organic queries will offer higher rewards than others.

This both makes it easier for researchers to work with us, and allows us to productize subnet 25. The choice to incorporate organic queries and a global job pool opens up a symbiotic relationship where miners can work with organizations that set jobs, allowing them to contribute to their goals.

Academic and professional milestones

Subnet 25 holds a unique position in the Bittensor ecosystem. As the first subnet focused on academia, it acts as the bridge between the network and the scientific community. Since its creation, we’ve been serving the Bittensor community with our dashboard, stats, and protein-folding releases. But now it’s time for SN25 to begin directly contributing to STEM. So keep an eye out for updates to our subnet, and announcements of new collaborations and partnerships - and proof that Bittensor can contribute to the advancement of one of the most significant scientific challenges of our era.