AI drug discovery stops being hypothetical
AI drug discovery is the use of computational drug design systems that combine large-scale biological data, machine learning, and automated experimentation to identify new therapeutic targets, design candidate molecules, and iteratively optimize them for safety and efficacy before advancing into human clinical trials. The most important shift today is that these AI-designed therapeutics are no longer marketing slides or proof‑of‑concept demos; they are clearing regulators and entering human studies. Recursion’s PI3Kα H1047R inhibitor REC-7735, an AI-designed PI3Kα inhibitor, has received IND clearance and will enter a phase I/II trial in patients with select PIK3CA H1047R-mutant solid tumors in the second half of the year. That milestone matters because it turns “clinical trials AI” from aspiration into a regulated reality, with all the scrutiny that implies.
Closed-loop AI workflows are redefining how targets are found
The most telling signal of maturation is that AI is now wired into continuous, closed-loop discovery workflows rather than bolted on as an afterthought. Recursion’s collaboration with a major pharma partner shows what this looks like in neuroscience: an end-to-end Data Factory built over more than a decade generates biological data explicitly designed for AI models. Researchers applied whole-genome CRISPR-Cas9 knockouts across more than 17,000 genes and thousands of small molecules to find gene–compound and gene–gene interactions, then advanced only those targets that cleared pathway, functional, and disease validation stages into a formal validation package. According to one collaboration update, it took 15 months to move from initiation of target validation to a package good enough for option exercise. That speed in a field where “only one in 40 neuroscience drugs to ever reach the clinic have been approved” is a shot across the bow for traditional playbooks.
From static assays to design–make–test–learn cycles
If the first AI wave focused on spotting targets, the next wave is about compressing the grind of medicinal chemistry. Receptor.AI and Sethera are explicit about this: they are building a closed-loop drug discovery workflow that ties physics-based modeling, artificial intelligence, and multiparameter optimization directly to experimental peptide libraries. Sethera’s encoded libraries install one to six cross-links to generate polymacrocyclic, nested, in-line, and interpeptide structures, exploring multiple experimentally accessible topologies instead of locking in a single motif. Receptor.AI interprets sequence, architecture, enrichment, and activity data to develop binding hypotheses, prioritize series, then guide focused optimization cycles. Subsequently, the companies will design, synthesize, and test new candidates, feeding each round of data into the next cycle. Together, they intend to move “more efficiently from experimental discovery to validated lead series,” a blunt admission that static enrichment-led workflows are no longer enough.
Why now: new tools for stubborn biology
The timing of this first clinical wave is not an accident; it reflects new tools colliding with areas where conventional discovery has stalled. In neuroscience, where more than three billion people live with neurological conditions yet only one in 40 drugs that reach the clinic is approved, Recursion and its partner built one of the first whole-genome CRISPR knockout phenomaps in living human neurons. They generated this map from a subset of over one trillion internally manufactured neuronal cells derived from induced pluripotent stem cells. Those new tools for the mostly unexplored neuronal genome are the foundation for the first validated neuro target discovered through their AI map. Meanwhile, Sethera’s polymacrocyclic architecture platform deliberately explores topologies that “conventional design approaches do not readily reach,” letting the biology and screening data, not human preconceptions, pick the winning shapes.
What comes next: from targets to pipelines, not one-off headlines
The uncomfortable truth for skeptics is that this is no longer about a single AI-designed molecule making headlines; it is about pipelines being rewired. For the neuroscience collaboration, the next steps include advancing the discovery program from a target into a drug by using Recursion’s chemistry platform to design a potential first-in-class molecule, while also identifying and validating additional targets. The neuro-focused maps are just two of six whole genome phenomaps; the other four aim to discover targets for an undisclosed GI-oncology indication. On the peptide side, Receptor.AI and Sethera plan, after validating their initial workflow, to pursue additional internal programs and joint discovery collaborations across selected target classes and therapeutic areas. This is the real meaning of REC-7735’s IND clearance: once AI pipelines prove they can repeatedly hand regulators viable candidates, the default assumption in drug R&D shifts from “why use AI?” to “why not?”
