Asimov Press has published a step-by-step technical guide on designing antibodies from scratch using AI tools — work that, by author Brian Naughton's own account, was not reliably possible twelve months ago.
Naughton, a Stanford-trained biomedical informatician and co-founder and CTO of cancer startup Decade, walks scientists through five stages: target selection, structure preparation using <a href="/news/2026-03-14-ai-engineer-uses-chatgpt-and-alphafold-to-develop-cancer-vaccine-for-his-dog">AlphaFold 3</a> and PyMOL, running design campaigns on cloud platforms including Ariax, Tamarind Bio, and Modal, filtering candidates by predicted binding affinity, and experimental validation. The pipeline lets wet-lab scientists screen antibody candidates computationally before any lab work begins — a shift that compresses early-stage drug discovery timelines without needing specialized infrastructure.
The guide centers on BoltzGen, an open-source tool developed by Hannes Stärk and the Boltz team — the same group behind Boltz-2, a structure prediction model that competes directly with AlphaFold 3. Released under the MIT license, BoltzGen is free for commercial use. The Boltz team reports sub-micromolar binding affinity in the majority of test cases across targets ranging from insulin to harder, less-characterized proteins. For scientists who want full antibody designs rather than fragments, RFantibody and the permissively licensed Germinal are alternatives Naughton covers in the guide. On the commercial side, Nabla Bio, Chai Discovery, Latent Labs, and Isomorphic Labs — the DeepMind spinoff — have all reported high success rates in de novo design.
Naughton uses Nipah virus Glycoprotein G as the guide's worked example: a surface protein on a pathogen with a 40–75 percent mortality rate and an existing antibody candidate already in Phase I clinical trials. His benchmark data comes from a design competition hosted by Adaptyv Bio, which screened more than 10,000 designs across tools including BindCraft, Mosaic, and BoltzGen — one of the largest public comparisons of AI antibody design approaches to date. Naughton was previously a founding scientist at 23andMe and maintains open-source repositories that package bioinformatics tooling on the same cloud infrastructure he recommends in the guide.
The Adaptyv Bio dataset will likely become a reference point for labs evaluating which tools to use. Naughton said he expects the number of viable open-source options to grow significantly over the next year.