Origin is using AI to accelerate drug discovery by predicting molecular behavior
Oct 8, 2025 with Yash Rathod
Key Points
- Origin's Axis model claims to outperform Google DeepMind's AlphaGenome by unifying multiple biological modalities in a single base model, departing from the narrow single-task approach that dominates biological AI.
- Origin plans to own its therapeutic pipeline rather than license its model, targeting a disease-specific gene therapy program within one year to avoid the AlphaFold trap of scientific prestige without commercial returns.
- Regulatory bottlenecks remain the hard constraint: even perfect simulations require animal and human trials, though Origin sees FDA openness to replacing animal toxicology studies as a positive early signal.
Summary
Origin is a San Francisco-based startup building AI systems to design drugs for complex diseases. Co-founder and CEO Yash used the segment to announce Axis, the company's first model, which he says outperforms Google DeepMind's AlphaGenome on its benchmark.
The technical bet behind Axis is unification. Most biological AI models are narrow, trained to do one specific task. Origin's argument is that biology demands a different approach — cells involve many interacting systems, and a model that can only predict one thing at a time can't capture that complexity. Axis is designed to handle multiple biological modalities within a single base model, which Origin claims is a first.
Near-term roadmap
Origin's immediate focus is gene therapy design — making therapies safer and improving efficacy. Yash says the plan is to expand Axis's capabilities across the components involved in designing those therapies, run the sequences the model generates through a wet lab, and have a therapeutic program targeting specific diseases within one year. The intent is to develop and own the therapeutic pipeline, not just license the model.
The Alphafold problem
The sharpest question Origin faces is the one that haunts every biology AI company: the gap between impressive science and commercial value. AlphaFold won a Nobel Prize and was eventually open-sourced, but never became a revenue-generating enterprise software business. Benchling, by contrast, sells electronic lab notebooks to biotech companies and charges for it. Origin's answer is that it isn't positioning as a foundation model lab hoping a product emerges — it intends to close the loop itself, using Axis to design therapies it will take through clinical development.
FDA and the testing bottleneck
The unavoidable constraint is regulatory. No matter how good the simulation, drugs still need to move through animal and human trials before approval. Yash points to the FDA already signaling a shift away from animal toxicology studies for monoclonal antibodies as an early positive indicator. His view is that regulatory frameworks will evolve in step with the technology — but that requires deep learning systems capable of recapitulating what happens inside tissues and eventually entire organisms, which is still ahead.
Origin is pre-revenue and early-stage, with no funding figures disclosed in this segment.