Research

Variant and diplotype interpretations are grounded in CPIC and DPWG guidelines and PharmGKB annotations. Gene–drug responses include star-allele context, phenotype assignments, and citations to the source guideline so you can verify every claim.

aafinomics generates ready-to-run R and Python pipelines for CUT&Run, CUT&Tag, ChIP-seq, and WGS data. Upload VCF, FASTQ, SAM, BED, or narrowPeak files, preview and sort columns in the Genomic Lab, then hand the sorted excerpt to the agent for analysis planning.

The retrieval layer queries public resources at request time: NCBI, TCGA via cBioPortal, NCI GDC (mutations, copy number, methylation, RNA expression, clinical data), PDB, UniProt, GTEx, Human Protein Atlas, AlphaFold structures, AlphaMissense pathogenicity scores, and KEGG pathways. Soft network failures degrade gracefully — a missing source is reported, not invented.

Unknown variants are dispatched to Arc Institute's Evo 2 biological foundation model running on TRUBA ARF and EuroHPC-class GPU nodes. Known pharmacogene variants are served from cache; novel sequences are scored as ΔLL = log P(alt) − log P(ref) by a persistent worker that keeps model weights pinned in VRAM across an allocation window.

aafinomics assists and streamlines research workflows — it does not replace established analysis pipelines or clinical review. Prognostic and survival outputs are research use only and are not clinically validated.