NOVA
SN68Uses AI to accelerate drug discovery, finding new medicines faster
NOVA turns drug discovery into a live, decentralized contest. Miners compete to surface viable drug candidates from a 65 billion compound chemical space, with weekly protein targets refreshed by the Metanova team. The repo updated two days ago, the most recent target is an IL-6 nanobody, and the alpha is up 55.76% over the past 30 days.
// Mining for medicine.
NOVA is an AI-powered drug discovery network. Miners compete each week to find molecules that bind tightly to a protein target chosen by the team, and validators re-score every submission using independent structural and binding models before paying rewards.
The simple version: Imagine a global treasure hunt where the prize is a molecule that could become a medicine. Every week the team picks a new disease-related protein. Thousands of search strategies run in parallel, and the cleanest hits get paid.
Centralized equivalent: Think Schrödinger or Recursion Pharmaceuticals, but the search engine is distributed across independent miners running their own models, with the scoring rules and target list published openly.
How it works:
- Miners compete in two tracks. In NOVA Compound, they submit candidate molecules tuned for high affinity to the weekly target, low affinity to anti-targets, and chemical novelty. In NOVA Blueprint, they instead submit search code that runs inside a standardized sandbox on an RTX 4090 against randomized targets.
- Validators re-score Compound submissions with Boltz-2 for structural fit and affinity, apply entropy bonuses for diverse chemistry, and invalidate any molecule already found earlier in the same target-week. For Blueprint, validators run the miner code, then re-score its 100-molecule output with PSICHIC.
- The problem it solves: Conventional drug discovery takes a decade per candidate and costs billions, with the search bottlenecked behind closed corporate pipelines. NOVA opens the early-stage molecular search to anyone with compute and a strategy.
- The opportunity: Pharma spends over 200 billion dollars a year on R&D. Even modest gains in the hit-finding stage compound into faster, cheaper preclinical pipelines.
- The Bittensor advantage: Decentralized competition means many independent search heuristics run side by side. Duplicate-invalidation rules push miners away from known scaffolds and toward chemistry no centralized lab has bothered to try.
- Traction signals: 583 commits and active weekly target rotations on the public repo, with the last commit landing two days ago. Recent work includes IL-6 nanobody target setup and enforcement of VHH hallmark residues at framework positions 49 and 50, indicating the team is tuning the scoring rules to real antibody biology rather than generic small molecules.
Category: Healthcare and Medical AI | Centralized Competitor: Schrödinger, Recursion Pharmaceuticals, Insilico Medicine
NOVA is one of the rare Bittensor subnets pointed at a tangible non-AI deliverable. Most subnets sell inference, training, or trading signals. NOVA is searching for actual drug leads against named protein targets, and it publishes the rules of that search in the open.
Mechanism:
The dual track design is what separates NOVA from a generic compute subnet. NOVA Compound is the direct competition: miners submit molecule sets each epoch, scored on target affinity, anti-target avoidance, and chemical novelty. A time-dependent entropy bonus rewards new scaffolds, and any molecule already found earlier in the same target-week is automatically invalidated, forcing continuous fresh exploration rather than miners converging on a few known winners.
NOVA Blueprint is the meta-competition. Instead of submitting molecules, miners submit search code. That code runs in a standardized sandbox for roughly 30 minutes on an RTX 4090, against randomized target and anti-target proteins revealed only after submissions close. The output, a set of 100 molecules, is then re-scored with PSICHIC. Intra-submission duplicates or molecules failing basic chemical-property checks (minimum heavy atom count, rotatable bond limits) disqualify the entire submission. The result is that NOVA pays not just for molecules, but for general-purpose molecule-finding tools.
The development picture is concentrated but steady. The public repo at metanova-labs/nova shows 583 total commits with 3 active contributors, led by cosmic-amanita. The most recent commit landed on 2026-05-20, two days before this write-up. The May commit history reads as ongoing biology work: MSA files added for proteins O15379, P05231, and Q6P6W3, a merged pull request enforcing VHH nanobody hallmark residues, a nanobody incentive proportion update, and weekly target refreshes. This is not a snapshot frozen in 2025, it is a live scoring engine being tuned each week.
Market metrics line up with that activity. NOVA trades at 0.02548 TAO for a market cap of about 123,342 TAO, with 44,992 TAO of root liquidity in the pool and a root proportion of 0.16, meaning roughly 84% of the pool comes from organic alpha demand rather than protocol subsidy. The 7-day net inflow of 945 TAO is solidly positive even after a slight 140 TAO outflow over the past 24 hours. The alpha is up 15.02% over 7 days, 55.76% over 30 days, and 103.63% over 90 days. The chain buy rate sits at 19.91%, with an emission EMA share of 10.69% under Taoflow. Concentration is moderate, with a normalized HHI of 0.046 and the top 10 stake positions controlling about 41% of stake.
The critical question is the same one facing every computational drug platform: do the molecules NOVA surfaces ever translate to validated wet-lab hits or preclinical candidates? The infrastructure is genuinely impressive and the team has pharmaceutical domain expertise that most crypto projects lack, with Micaela Bazo as CEO, Pedro Penna as CSO, and Amanda Casadei as CTO. But the value of any in-silico screen is ultimately decided in a lab. That is a step NOVA itself cannot perform, and the team has not publicly announced a partnership moving a NOVA-discovered molecule into preclinical testing yet.
- Validation gap: Computational screening is only the first step in drug discovery. Without wet-lab partnerships to test candidates and a track record of advanced compounds, NOVA's outputs remain theoretical.
- Regulatory complexity: Drug development requires navigating FDA and EMA processes that no subnet can automate. The path from a high-affinity prediction to an approved medicine is long, slow, and capital-intensive.
- Niche audience: Pharmaceutical AI is a specialized market. Most crypto miners and TAO stakers may not recognize the value proposition, which can cap how widely the subnet is understood and held.
- Contributor concentration: Public commit activity is currently driven by a small core team, with one developer responsible for the majority of recent work. Slower bus factor on a research-heavy codebase is something to monitor.
- Competition from well-funded incumbents: Schrödinger, Recursion, Insilico Medicine and others have raised hundreds of millions for AI drug discovery, with established pharma partnerships and proprietary datasets NOVA cannot match yet.
Into the next one.
