Decentralizing drug discovery: protein docking on SN25
In order to impact medicine, life sciences on the blockchain must incorporate a wider range of the drug discovery process. That's why SN25 now supports protein docking.
Understanding how proteins behave is key to drug discovery. But to make progress towards real-world impact, we need more than just modelling of how they fold. Scientists working in drug discovery need to know how the proteins bind to molecules in order to function best.
That’s why we’re introducing a new task, docking, to SN25. As well as expanding the subnet’s value to pharmaceuticals research, it’s also a step towards tackling a wider range of scientific research problems - which is why we renamed SN25 to Mainframe.
By proving our value through the drug discovery vertical, we can prove the subnet’s commercial applicability while also extending our sphere of activity into a wider world of decentralized science. In turn, this can demonstrate Bittensor’s potential to create social impact and push our protocol into critical areas of scientific research.
To achieve that, however, we must expedite drug discovery by implementing docking. But what exactly is it, and why’s it so important?
A primer on docking
Say you’re trying to build a new cancer treatment. Like most drugs, yours is made up of several connected molecules (such as proteins and ligands). You’ve figured out which molecules you’re working with, but you’re struggling to know how they should connect. There are several ways they could theoretically bind, but not all bindings are equal. Some generate unwanted effects, while others could be unstable, meaning the molecules will slip out of place and drift away.
The placement of each molecule is just as important as selecting the right molecules. This is easiest explained by an old saying in biology - ‘form defines function’. The way something’s structured determines how it behaves.
When discovering new drugs, you must observe how the molecules involved attach to each other, as this changes how the final molecule behaves. You must also discern how they interact with their surroundings. In the words of SN25’s subnet lead, Brian McCrindle, ‘Proteins and other molecules are not static objects. These are things that evolve over time. They change shape and react differently depending on their environment’.
This is what docking is all about. It’s the process of figuring out which molecules connect and where they connect best to each other, so they can perform a stable and meaningful function. Docking involves computational tests and machine learning inference to predict where molecules can bind.
Docking is a major part of the AI drug discovery sector, which is why it’s being incorporated into SN25. ‘We want to continue to build tools that allow researchers to have ongoing confidence in what they’re creating,’ Brian explains. Pharmacologists never use just one app, but rather a bundle, with each one helping add to their findings and refine their experiments. SN25 is focused on being a part of that pantheon of tools. We’re creating ‘common good’ solutions to help push drug discovery forward in an open, transparent, and accessible way.
Docking on SN25
To perform docking on SN25, we’ll be running machine learning inference on the validator-side. This is powered by DiffDock, an AI model for molecular docking. It lets you predict how proteins and ligands will bind, testing different environmental conditions, and checking a range of locations on both objects. A ligand is a molecule that binds to a protein, often affecting its function.
There are other objects that can be docked to each other, but for SN25, we’re currently focused on protein and ligand bindings. This relationship is the backbone of drug design, and is essential in developing life-saving treatments. In fact, some essential medications - such as Imatinib, a leukemia treatment, and Ivacaftor for cystic fibrosis - result from the binding of just one ligand and one protein.
Docking and protein folding
You might read this and wonder if SN25 is still doing molecular dynamics (MD), or if it’s switching over fully to docking. The answer is we’re doing both. We’re not stopping with our work in MD - we’re simply adding more functionality to the subnet.
MD will run alongside docking. They both serve separate, but deeply connected functions. One is about producing proteins, and the other is about binding ligands to proteins. Together, they help build an essential part of the drug design pipeline.
Researchers or companies can come to SN25, post a specific molecular dynamics task in our global job pool, get the miners to simulate the dynamics of their desired proteins, and then use the subnet to discover the optimal docking solutions - giving them a more complete service in their drug discovery process.
Bittensor’s venture into DeSci
At the heart of SN25, Mainframe, is the desire to strengthen the scientific community. We’re empowering researchers and academics, offering them decentralized solutions to their hardest problems. We’re crowdsourcing computing power across the globe and directing it towards life sciences. As a result, this can enable Bittensor to make meaningful contributions to science.
As Brian explains, ‘for SN25, we continue to push what we can provide researchers with by being able to ask important questions. For instance, helping researchers learn how particular proteins and ligands interact with each other and how they evolve in particular environments’.
To learn more about docking, protein folding and Mainframe, come and chat in our Discord and Telegram.