Codebreaker Labs builds the causal data layer for genomics AI by editing single-nucleotide variants into primary human cells and measuring their functional effects. Its CODEX platform generates reference atlases that link genetic variants to disease-relevant cellular outcomes, providing the ground truth needed to train more accurate genomic models.
Funding
Funding not disclosed
Founders
Product
Problem
Genome-wide association studies (GWAS) and whole-genome sequencing (WGS) identify variants associated with disease, but they do not reveal which variants are causally responsible or what they do in cells. Existing functional genomics approaches rely on gene-level knockouts in immortalized cell lines, which operate at a coarser resolution than the single-nucleotide variants patients actually carry and fail to reflect disease-relevant primary cell biology.
Solution
Codebreaker Labs operates CODEX, a causal data layer for genomics AI that resolves genetic variation to single-base causal resolution. The platform uses multiplexed CRISPR editing to introduce thousands of specific variants in parallel into disease-relevant primary human cells, then measures the functional consequences of each edit. Each completed project produces an Atlas—a reference dataset linking variants to cellular function—along with AI models trained on that data. These atlases are built through a designed pipeline that combines in-house AI models that nominate variants worth testing, multiplexed editing in validated cell-type platforms, and optional collaboration with key opinion leaders for clinical direction. The resulting data and models are licensable assets that compound as more atlases are completed, providing the training ground for next-generation genomic AI.
Target Audience
Primary customers are pharmaceutical and biotech companies, genomics AI developers, and clinical research organizations that need causal variant-function data to train genomic models, validate drug targets, or interpret patient genomes.
Features
- Single-nucleotide causal resolution, testing variants one at a time at scale rather than relying on regional associations or gene-level knockouts
- Multiplexed CRISPR editing in primary human cells, with up to 10,000 variants edited in parallel per program
- Cell-type platform lanes (e.g., T-cell platforms for immunology and oncology indications) that enable immediate program starts within validated systems
- In-house AI models that mine public and partner datasets—including GWAS Catalog, ClinVar, biobank cohorts, and published case literature—to nominate every variant worth testing
- Atlases delivered as finished, licensable assets combining measured data with trained models, with the option to license data, models, or both
- Pipeline design that breaks linkage disequilibrium by testing variants individually in disease-relevant primary cells rather than immortalized lines