NOVA Takes AI Drug Discovery From Competition to the Lab

NOVA has reached a new stage in its attempt to turn drug discovery into a continuous, open competition. Its first small molecules and nanobodies have entered the laboratory for experimental testing.

This development marks an important milestone for the subnet. Until now, NOVA’s competitions have primarily operated in the computational domain, with participants submitting molecules, nanobodies, and discovery algorithms and competing based on measurable performance.

With physical experiments now underway, the network can begin bringing real-world experimental results back into the system.

212,000+ Submissions

The scale of the competition is already substantiated. NOVA reported that it has received 212,561 submissions, enriching its discovery libraries with:

• 11,130,975 small molecules
• 82,575 nanobodies
• 6,397 discovery algorithms

These submissions come from three parallel competitions covering small-molecule design, nanobody design, and chemical search algorithms.

This is significant because the competition creates a mechanism for continuously selecting among different approaches without relying on a single model, research team, or predetermined methodology.

Now, some of those computationally generated candidates are being subjected to experimental validation.

From Benchmark Performance to Physical Ground Truth

One of the persistent problems in AI-driven drug discovery is the gap between performance on computational benchmarks and performance against novel, real-world chemistry.

NOVA’s approach is designed to address that gap by introducing experimental results into the competitive loop.

The subnet’s development builds on a broader history of scientific competitions, from Kaggle’s Merck Molecular Activity Challenge and Tox21 to Leash Bio’s BELKA challenge and more recent blind and prospective drug-discovery benchmarks.

These competitions have demonstrated the value of exposing difficult scientific problems to large numbers of competing approaches. But NOVA is attempting to make the process persistent, while extending the evaluation process toward physical experimentation.

A Continuous Discovery Loop

This is where NOVA’s model becomes particularly interesting.

Instead of a competition ending when a leaderboard is finalized, successful computational candidates can move toward laboratory testing. The resulting experimental data can then provide new ground truth for the network.

That creates the possibility of a continuously improving discovery system and, in turn, determines which ideas deserve further resources.

The platform is effectively trying to turn drug discovery into an ongoing selection process. If that loop works at scale, the value of NOVA may extend well beyond the individual competitions they’re currently running.

The objective is no longer simply to find the best-performing model on a benchmark. It is to build an infrastructure where the best-performing scientific ideas can continuously move from computation toward experimentally validated discoveries.

And with the first small molecules and nanobodies now in the lab, NOVA is beginning to test that objective against physical reality.

More on Metanova (SN68):

Hash Rate Ep. 165: Metanova Expands Into Nanobodies, Wet Lab Pipeline, and AI Co-Scientist Vision