alterlab-borzoi

# Borzoi (sequence → function) ## Overview **Borzoi** (Linder et al. 2025; `calico/borzoi`) is a **sequence-to-function** deep-learning model: given a DNA sequence over a long genomic context, it predicts **genome-wide functional tracks** — RNA-seq, CAGE, ATAC-seq, and ChIP coverage across many assays/tissues. Its headline use is **non-coding variant effect scoring**: run the reference and alternate alleles through the model and compare predicted tracks to estimate a regulatory variant's impact on expression or chromatin. It **predicts** function from sequence; it does not *look up* known variants. For a variant's population frequency use `alterlab-gnomad`; for clinical significance use `alterlab-clinvar`. ## When to Use This Skill Use this skill when the user wants to: - Predict **functional tracks** (RNA-seq/CAGE/ATAC/ChIP) from a DNA sequence or locus. - Score a **non-coding / regulatory variant's** predicted effect (ref vs. alt). - Run **in-silico mutagenesis** to find driver bases in a regulatory element. - Prioritize candidate regulatory variants by predicted functional impact. ### Does NOT Trigger | Scenario | Use instead | |----------|-------------| | Look up a variant's **population frequency** | `alterlab-gnomad` | | Look up a variant's **clinical significance** | `alterlab-clinvar` | | Predict a **protein-structure** / coding effect | `alterlab-alphafold` | | Single-cell foundation-model tasks | `alterlab-scgpt` | | Standard variant calling from reads | `alt...