AI drug discovery now means targets that stand up in the lab
AI drug discovery is the use of computational models and large biological datasets to identify, prioritize, and experimentally validate new drug targets and molecules, aiming to transform trial‑and‑error research into a systematic and data‑driven process that can scale across difficult therapeutic areas. Today, the real story is that AI‑designed drug targets are no longer a science‑fiction concept or a slideware promise; they are beginning to survive the harshest test in pharma: independent validation by large partners who are willing to build drug programs around them. The most telling proof is in neuroscience, where traditional approaches have failed so often that only a tiny fraction of candidates have ever made it from Phase I studies to approval. If AI can change the odds in this field, the rest of drug development will have to follow.
Recursion and Genentech: an AI map delivers a first validated neuro target
The clearest sign that AI‑discovered targets are maturing is the decision by a major pharma group to pick up an option on a neuroscience program based on an AI‑generated map of neuronal biology. In this collaboration, Recursion and Genentech are advancing a small‑molecule discovery effort around what they call the first validated neuroscience target discovered through an AI map. The companies built a whole‑genome CRISPR knockout map from more than a trillion induced pluripotent stem cell–derived neurons, giving them a broad view of gene‑gene and gene‑compound interactions in a human neuronal context. Instead of focusing on a handful of familiar pathways, they scanned the mostly unexplored genome in living neurons to surface novel biology. The key point is not the secrecy around the target itself, but that Genentech accepted a rigorous, jointly built validation package, signaling confidence that this AI‑designed drug target is worth a full chemistry campaign.
Closed-loop AI workflows: from static models to learning systems
While Recursion’s neuro map demonstrates AI’s power to find targets, Receptor.AI and Sethera show where the field is heading: closed‑loop, experiment‑driven learning systems for difficult therapeutic targets. Their partnership centers on a workflow in which Sethera generates and screens architecture‑diverse polymacrocyclic peptide libraries, and Receptor.AI applies physics‑based modeling, artificial intelligence, and multiparameter optimization to interpret sequence, architecture, enrichment, and activity data. Crucially, they commit to an iterative loop: design, synthesize, and experimentally test new candidates, then feed that data back to guide the next cycle. As Receptor.AI’s CEO puts it, the goal is to use physics and AI not as a substitute for experimentation, but to learn from each experimental cycle and direct the next one. This mindset treats AI as an active partner in the lab, not a one‑off prediction engine.
Why partnerships matter more than algorithms
The common thread in these efforts is not a specific AI architecture but a partnership structure that forces AI to meet experimental reality. Recursion’s work depends on an end‑to‑end data factory that produces neuronal phenotypes at a scale outside cell manufacturers initially considered practical, backed by a jointly defined validation cascade of pathway, functional, and disease‑level studies before any target earns a formal package. Receptor.AI and Sethera are similarly explicit: Sethera’s chemistry platform can access peptide architectures that conventional design approaches do not reach, and AI is judged on how well it improves hit confirmation, selectivity, and lead quality compared with conventional workflows. In both cases, the value lies in tying AI to platforms that generate new data, so models are constantly tested and refined rather than frozen in time.
From validated targets to full pipelines
The most important question now is what happens after a target survives validation. Recursion and Genentech plan to advance their neuroscience program from a target into a drug using Recursion’s chemistry platform to design a potential first‑in‑class molecule and to identify and validate additional targets for new programs. Recursion also intends to keep combining its phenomics dataset with Genentech’s proprietary transcriptomics data to build more multimodal maps that link gene perturbations to cellular phenotypes in search of further novel pathways. On the peptide side, Receptor.AI and Sethera expect that, once they prove the closed‑loop workflow on an initial hard‑to‑drug target, they will extend it to additional internal projects and jointly structured discovery collaborations with other pharmaceutical and biotechnology partners. AI drug discovery is not ending at target lists; it is starting to generate pipelines that will live or die on experimental and, ultimately, clinical outcomes.
