Finding signal on Twitter is more difficult than it used to be. We curate the best tweets on topics like AI, startups, and product development every weekday so you can focus on what matters.
Originally published by @twistartups on X. Tech Twitter preserves the original source alongside this readable edition.
Before you dive in
• Bittensor subnet MetaNova disrupts drug discovery with open, decentralized competition—65B molecule search space, 5.5M+ submissions, faster innovation than...
Best for builders who want practical takeaways. 7 min read.
Getting a single drug approved, on average, takes 10 years and $2.6B. Despite all the time and money, more than 90% of drug candidates still fail clinical trials. This is what the industry calls a “pipeline crisis”. Everyone agrees it needs fixing. Nobody inside the system has fixed it.
That's where it gets interesting. Because a small team operating out of Lima, Peru — running a subnet on a crypto network most people have never heard of — might be closer to a solution than anyone in Big Pharma's boardroom would like to admit.
Hold on... What's Bittensor?
If you've never heard of Bittensor, here's what you need to know. Bittensor is a crypto project with the token, $TAO. It uses crypto incentives to pay people, and increasingly AI agents, to produce useful AI outputs. Think Bitcoin, but instead of burning electricity to validate transactions, contributors compete to solve real AI problems.
The network is organized into 128 “subnets,” each focusing on a different problem. Each of these subnets have their own tokens (called alpha tokens), their own rules, and their own rewards. Like Bitcoin, TAO has a hard cap of 21 million tokens.
The people who compete are called miners. The people who judge and score their work are called validators. That's the system.
The problem with how we find drugs
Before a drug can be tested on animals or humans, researchers have to find a molecule worth testing. That process involves searching through an enormous library of compounds to find ones that might bind to a specific target.
The challenge is scale. The theoretical chemical universe contains somewhere around 10⁶⁰ possible drug-like molecules. No lab, no matter how well-funded, can search more than a tiny proprietary slice of that space. Most searches use the same well-known databases, the same established techniques, and the same internal models. The industry is, in effect, looking for its keys under the same lamppost over and over.
There are structural issues at play here too. Drug discovery is centralized, and gatekept. Big Pharma labs protect their methods. Breakthroughs sit behind patents. Academic institutions publish results slowly and selectively. There's no mechanism for a random engineer in another country to contribute a better idea. There's no incentive for anyone outside the system to try.
In comes MetaNova...
We doomscroll, you upskill
Get the 10 tweets shaping how builders think today.
Newsletter
We doomscroll, you upskill
Get the 10 tweets shaping how builders think today.
MetaNova Labs, the team behind Bittensor Subnet 68, launched in March 2025 with a simple but radical premise: what if drug discovery was an open, 24/7 global competition?
Here's how it works. MetaNova sets a target. For example, find the most promising molecules that bind to a specific receptor involved in mood or reward. That challenge goes live on the subnet. Miners from anywhere in the world — engineers, researchers, AI agents — compete to submit the best candidates. Validators score the submissions using MetaNova's proprietary scoring model, which predicts binding affinity. The best submissions earn $NOVA tokens. Then the next challenge starts immediately.
“It's a hackathon that never sleeps — applied to one of the hardest problems in science.”
The search space they're working with is staggering. MetaNova started with a database of 1.75 billion synthesizable molecules. Then they layered on combinatorial chemistry — essentially simulating how molecules can be assembled from real chemical reactions in a lab — expanding the searchable universe to an estimated 65 billion possibilities. Every molecule in that pool is real and synthesizable. This isn't theoretical noise. These are compounds that can actually be made and tested.
What the subnet does that the industry can't
The most important distinction isn't the technology. It's the incentive structure.
Traditional pharma
Centralized & gatekept
Methods hidden behind IP
Same databases, same tools
Closed to outside ideas
Slow feedback loops
Geographic limitations
MetaNova subnet
Open to anyone, globally
All results on-chain
65B molecule search space
Cross-disciplinary miners
Continuous improvement
Target-agnostic platform
In a traditional pharma lab, if an AI model has a blind spot — an area where its predictions are systematically overconfident — that weakness stays hidden until it causes a failure deep in the pipeline, after years of work and hundreds of millions of dollars. On the MetaNova subnet, miners immediately try to exploit any weakness in the scoring system to win more tokens. That adversarial pressure surfaces model failures fast, in the open, where they can be fixed. What looks like a bug is actually one of the subnet's most powerful features.
The other thing the subnet does that no pharma company can replicate is attract genuine cross-disciplinary insight. Because the problem has been reduced to a search and optimization challenge, contributors don't need a biology background. They need to be good at search algorithms. That opened the door to something remarkable: a miner with no pharmaceutical training applied an optimization strategy that had never before been used in drug discovery — and outperformed a well-established industry technique across multiple protein targets. That kind of cross-pollination is structurally impossible inside a closed research institution.
The results so far
MetaNova has been live for just over a year. The numbers are already meaningful.
5.5M+
molecule submissions to the network
8,700+
unique protein targets explored
260
active global contributors at peak
More importantly, it's stopped being theoretical. MetaNova has partnered with Yalotain, a Shanghai-based biotech, to move 50 of the subnet's top candidate molecules into wet lab validation — the stage where you actually synthesize the compound and test whether it does what the model predicted. This is the bridge between the computational world and the real one, and crossing it is a significant milestone for any drug discovery program, let alone a decentralized one.
The business model
MetaNova operates as what the industry calls a virtual biotech. They keep a lean team and no expensive internal labs, using contract research organizations (CROs) to handle synthesis and testing of the candidates the subnet surfaces. This keeps overhead low and execution fast.
From there, the revenue paths are multiple. Early-stage IP can be licensed to larger pharma companies. The platform itself can be offered as a screening service to other drug developers who want to run their own targets through the subnet. Assets can be co-developed with partners at various stages. And by conducting clinical work in jurisdictions with favorable regulatory agreements — the FDA has been expanding its acceptance of international trial data — the cost of validation can be compressed significantly further.
The key insight is that the subnet isn't just MetaNova's internal drug pipeline. It's infrastructure. Any biotech, any pharma company, any academic lab with a target they want screened could theoretically plug in. That's not a drug company with one or two shots on goal — that's a platform with unlimited shots.
Why this matters beyond crypto
It would be easy to file this under “interesting crypto experiment” and move on. That would be a mistake.
The structural problems MetaNova is attacking — opaque incentives, gatekept research, a system that financially punishes early failure rather than learning from it — are not problems unique to pharma. They're the problems of any centralized, high-stakes R&D process. What Bittensor's subnet architecture enables is a genuinely different approach: permissionless, continuously improving, and aligned so that the people doing the best work get paid the most, regardless of where they are or what credentials they hold.
Drug candidates from AI-assisted discovery are already in late-stage clinical trials. The era of AI-developed therapeutics reaching patients is not a decade away — it's probably three to five years. The question is which model of discovery gets there first: the $2.6 billion, decade-long, 90%-failure-rate incumbent, or an open network of global contributors running a nonstop hackathon on 65 billion molecules.
MetaNova is betting on the latter. And the early evidence suggests that bet isn't as crazy as it sounds
