{"entity":{"id":"enformer-borzoi","kind":"technology","name":"Enformer and Borzoi (DeepMind, Calico)","aka":[],"tldr":"Models that predict how DNA sequence controls gene activity, used to interpret non-coding cancer mutations.","summary":"Enformer and Borzoi are sequence-to-function models that predict how DNA controls gene activity by combining convolution with a transformer over long DNA windows. Enformer (Nature Methods 2021, DeepMind) predicts gene expression and chromatin signals from 200 kb of sequence; Borzoi (Nature Genetics 2025, Calico) extends this to predicting RNA-seq coverage and splicing across 500 kb. Both are used to interpret non-coding variants, including candidate cancer drivers in promoters and enhancers, by comparing predictions for reference and mutant sequence. Their cell-type coverage is bounded by the assays in the training data, so effects in tissues or tumour states that were never profiled cannot be predicted reliably. For a newcomer: these models read a long stretch of DNA and predict how a mutation there would change which genes are switched on.","status":"emerging","asOf":"2026-09-08","links":[{"label":"Enformer, Nature Methods 2021","url":"https://doi.org/10.1038/s41592-021-01252-x"},{"label":"Borzoi, Nature Genetics 2025","url":"https://doi.org/10.1038/s41588-024-02053-6"}],"tags":["foundation-model","genome"],"related":[],"cancers":[],"sections":["ai-computation"],"technologies":[],"targets":[],"drugs":[],"companies":["google-deepmind"],"institutions":[],"pathways":[],"terms":[],"trials":[],"people":[],"bottlenecks":[],"keyPapers":["paper-linder-nat-genet","paper-avsec-z-nat-methods"],"journals":[],"dependsOn":[],"notes":[],"principle":"Enformer and Borzoi combine convolution with a transformer over long DNA windows.","strengths":["Regulatory variant interpretation"],"limitations":["Cell-type coverage bounded by training assays"],"since":2021},"route":"/technologies/enformer-borzoi/","neighbours":{"section":[{"id":"ai-computation","kind":"section","name":"AI & Computation","route":"/fronts/ai-computation/"}],"company":[{"id":"google-deepmind","kind":"company","name":"Google DeepMind (and Google Research)","route":"/companies/google-deepmind/"}],"paper":[{"id":"paper-avsec-z-nat-methods","kind":"paper","name":"Effective gene expression prediction from sequence by integrating long-range interactions","route":"/key-papers/paper-avsec-z-nat-methods/"},{"id":"paper-linder-nat-genet","kind":"paper","name":"Predicting RNA-seq coverage from DNA sequence as a unifying model of gene regulation","route":"/key-papers/paper-linder-nat-genet/"}]}}