Fully unraveling the mysteries of biology is one of the most impactful technical challenges of our time and is only possible due to the current acceleration in AI.
We started this research lab to tackle this problem with a simple goal: unravel ALL biological mysteries. We, somewhat short-sightedly, call our lab GPTomics. While we use many transformers, that name no longer fits fully as achieving this goal will require a broad set of technologies, but we'll still use it to remind us of the start. We're taking a multi-path approach to understanding biology by both building new models to simulate biological systems at different scales, and customizing general-purpose tools (e.g., agents) to improve how we research and discover patterns in biology. Our plan is to move as fast as possible so humanity can benefit from what we learn.
We truly believe this is the moment and we are the generation that can unravel biology down to its core principles and build a reductionist foundation that holds up to the rigor of falsifiable science. We plan to build in public for as long as we can. If you have ideas, suggestions, or want to contribute, submit a PR or contact us.
Latest
Blog · The Wet Lab is Part of the Model
AI can propose the next experiment, but the cells still get the final vote. A wet lab view of why the assay, not the model, sets the ceiling on what a discovery loop can learn.
Blog · Relative To What
Genomics still depends on privileged coordinates. A look at what remains true when the reference genome becomes bookkeeping instead of foundation.
Research
BioJEPA
Applying JEPA-based architectures to biology: building an action-conditioned "world model" for cells inspired by V-JEPA 2-AC, aiming for a digital simulator that predicts how cell states respond to perturbations.
Genomic Relativity
Essay seriesA developing way to describe genomic positions and variants without treating any one reference assembly as the biological center.
COFR
Current Overview of Full Research helps AI coding agents maintain the live state of long-running research projects across files and sessions, tracking beliefs, evidence, decisions, open questions, and risks.
BioSkills
PausedA collection of skills that guide AI coding agents (Claude Code, Codex, Gemini) through common bioinformatics tasks, with code patterns, best practices, and examples across analyses from sequence work to single-cell and population genetics.
Bio-Task Bench
PausedBio-Task Bench is both a benchmark and evaluation tool for evaluating how well AI coding agents perform bioinformatics tasks. It includes 33 original tests across 10 domains, a deterministic grading harness, and adapters for running external bioinformatics benchmarks under the same CLI.
Contact
If you have something to tell us, email us at contact@gptomics.com, or on Twitter/X @gptomics.