BoltzProt-1 shifts the drug-discovery bottleneck from computation to the wet lab
BoltzProt-1 raises confirmed binder rates from 3.3% to 8% on difficult targets; with both models available via API, the bottleneck moves toward lab validation.
BoltzProt-1 designs proteins that bind to a given target, including nanobodies. On a panel of difficult targets, the share of confirmed binders rose from 3.3% for BoltzGen to 8% for BoltzProt-1. The hit rate is still low, but for lab design it changes the cost of each iteration.
BoltzMol-1 works differently: it takes a target protein and ranks the small molecules worth testing in the lab. Boltz's technical report says the system found active molecules or binders for 6 of 10 hard targets. The validation sets were small — dozens of compounds per target, in one case 96. Standard screening runs through tens of thousands to millions of compounds. Not industrial scale yet, but for a small team, ranking dozens of candidates beats screening millions.
Phylo integrated both models into Biomni Lab. An agent takes a text prompt, finds a target via Open Targets or single-cell data, sends it to Boltz, and receives predicted structures, confidence scores, and a ranked candidate list. In the examples, the agent selects IL23R as a target for ulcerative colitis, designs 50 nanobodies, and sorts them by predicted binding. Agentic biology here means chaining model calls in one pass — from target selection to a ranked list.
Boltz released an API to run both models. If a small team can go from target to first testable candidate with a batch of molecules and an API call instead of a massive screen, the barrier to entry drops. But without a lab experiment, a prediction is still an image and a number. Lab validation is mandatory.
In an NCBI Virus benchmark, science agents returned 106, 15, and 5 sequences instead of the correct 266 — until a dedicated database-access tool improved accuracy. Boltz plays a similar role on a different part of the pipeline: helping choose which molecules are worth testing.
Boltz's June 16 launch is the clearest recent illustration of the 'cheap compute, expensive validation' gap: the computational phase just became an API call, which immediately surfaces the next bottleneck — wet-lab budget. Experiment is a crowdfunding platform that funds precisely this kind of small-scale, specific, testable biology, making the Eternal Search organization page for Experiment the most directly relevant resource for readers who now have API-generated drug candidates but need bench funding.