Computational drug discovery startup Deep Origin announced in mid-August 2026 that its AI platform achieved a 31% hit rate identifying viable molecules against a cancer immunotherapy enzyme target, a figure the company says is nearly 100 times higher than what conventional machine-learning-only approaches typically deliver on similarly difficult targets. The claim, first reported by HPCwire’s AIwire vertical, centers on Deep Origin’s approach of pairing physics-based molecular simulation with AI models, including newer co-folding architectures that predict how a candidate drug molecule and its target protein bind together.
What a “hit rate” actually measures
In early-stage drug discovery, a “hit” is a candidate molecule that shows real biochemical activity against a target when tested, as opposed to the vast majority of computationally screened candidates that turn out to be duds once synthesized and assayed. Hit rates in traditional machine-learning virtual screening for hard enzyme targets often run under 1%, meaning researchers must synthesize and test hundreds of molecules to find one worth pursuing further. A 31% hit rate, if it holds up under independent replication, would mean dramatically fewer wasted synthesis-and-test cycles — the single biggest cost driver in early drug discovery.
How Deep Origin says it got there
Rather than relying purely on pattern-matching AI models trained on existing chemical databases, Deep Origin layers physics-based molecular dynamics simulations underneath its AI predictions, using the physics to constrain and validate what the AI model proposes. The company has also incorporated newer co-folding models — AI systems, related conceptually to AlphaFold-style protein structure prediction, that jointly model how a small molecule and a protein target fold together rather than treating protein structure as fixed. Deep Origin frames this hybrid physics-AI approach as differentiated from pure deep-learning drug discovery platforms that rely almost entirely on training data patterns.
Why the biotech industry is watching hit rates so closely right now
2026 has been described by trade outlet Drug Discovery News as an inflection point for AI in pharma, with roughly half of biotech companies using AI reporting faster time-to-target and about 42% reporting improved accuracy or hit rates from AI-assisted discovery work, and with 80% of organizations planning to increase AI budgets over the next year. Against that backdrop, the ultimate test the industry is waiting for is not a screening statistic but Phase III clinical trial results — proof that AI-discovered molecules actually work safely in patients, which no AI-native drug has yet fully delivered at scale.
Reasons for caution before celebrating the number
The 31% figure comes from Deep Origin itself, has not yet appeared in a peer-reviewed publication, and applies to one specific enzyme target rather than a broad panel of drug targets with varying structural complexity. Cancer immunotherapy enzymes are a category where structural biology is relatively well characterized compared with, say, intrinsically disordered proteins or membrane-bound receptors, so a strong result on one target does not necessarily generalize. Drug discovery veterans have grown wary of hit-rate claims in press releases after several AI drug discovery companies in prior years touted early virtual screening statistics that did not translate into approved drugs, or in some cases not even into successful Phase I trials.
The competitive landscape this sits inside
Deep Origin’s announcement lands amid a broader capital rush into AI-native biotech, exemplified by protein design company Chai Discovery’s reported valuation tripling to $3.8 billion on the strength of partnerships with OpenAI and Eli Lilly. Deep Origin is a smaller, less-capitalized player positioning its physics-plus-AI hybrid as a technical differentiator against both pure deep-learning shops and traditional computational chemistry vendors, in a market where investors are increasingly demanding evidence beyond in-silico screening statistics.
What’s next
Deep Origin has not disclosed a timeline for moving the cancer immunotherapy candidates identified through this screen into wet-lab synthesis, animal testing or human trials, each of which represents a much higher bar than a computational hit rate. Independent researchers and pharma partners will be watching for peer-reviewed publication of the methodology and for Deep Origin to demonstrate the same hit-rate advantage across additional, more structurally diverse targets. As with all AI drug discovery claims in 2026, the number that will ultimately matter is not the hit rate but whether any molecule identified this way survives the years-long path to an approved medicine.
