Borzoi predicts functional coverage tracks — RNA-seq, CAGE, DNase, ChIP — straight from DNA sequence. Reach for it to get predicted tracks across a locus, or to compare ref and alt windows around a variant when you want the assay-level consequence rather than a language model's likelihood; evo2 is the skill for the likelihood, and the two answer different halves of the same variant question. What is here is guidance for driving an external PyTorch port. No model runtime and no checkpoint is bundled.
Whether any of it runs depends on the environment: compatible packages, weights already downloaded, the track metadata, and a substantial amount of GPU memory. A predicted track delta is model evidence you can prioritize on. It is not causal proof, and it is not clinical validation.
A Skill is a directory of files, so installing one is copying that directory
somewhere an agent looks. With Node 18+ and git on PATH (npm resolves a
github: spec through it), and nothing cloned:
npx github:PKU-YuanGroup/OpenAI4S install borzoi --target claude
--target claude writes to ~/.claude/skills, claude-project to
./.claude/skills, openai4s to <data_dir>/user-skills, and --dir <path>
to anywhere you name. The resolved absolute path is printed before anything is
written there, and --dry-run stops at that plan. A reinstall refuses to
overwrite a copy you have edited or one it did not install, and uninstall
removes only the files it wrote. For a recipe included in the npm release,
npx @pku-yuangroup/openai4s-skills install borzoi --target claude installs the
published copy. The npm catalog can differ from this repository.
Without Node, take the directory itself — and turn it into a .zip if an upload
field wants one. The tarball is the whole repository (over 100 MB), and the pipe
is a POSIX shell recipe (macOS, Linux, WSL) that extracts only this directory:
curl -L https://codeload.github.com/PKU-YuanGroup/OpenAI4S/tar.gz/refs/heads/main \
| tar -xz --strip-components=2 OpenAI4S-main/skills/borzoi
python3 -m zipfile -c borzoi.zip borzoi
The click-through form of that same download is the
repository zip,
which is also the route to take on Windows: extract it and copy skills/borzoi/
out (unzip main.zip 'OpenAI4S-main/skills/borzoi/*' unpacks only that
directory). If you already run OpenAI4S there is nothing to install — the wheel
ships every bundled Skill, and a bundled Skill takes precedence over a
same-named copy in <data_dir>/user-skills. Targets, provenance, and what the
installer refuses to do:
tools/skills-installer/.
| File | Responsibility |
|---|---|
SKILL.md |
The input window is fixed at 524,288 bp of one-hot DNA and the model exposes no attribute that says so, so a shape mismatch is the first thing you meet: pad or crop. Out comes a tensor of 7,611 human tracks over 32-bp bins, with the separate 2,608-track mouse head off unless you enable it and select it. From there: where the track metadata actually lives (TRACKS_DF, not a targets attribute the base model does not have) and how to line it up with the output, ref/alt variant scoring, the VRAM floor, and the CC-BY-4.0 terms on the ported weights, which do not match the Apache-2.0 code they came with. |