name: borzoi_skill description: > Predict genome-wide functional tracks (RNA-seq, CAGE, DNase, ChIP) from DNA sequence with Borzoi. Use this skill when: (1) Scoring the regulatory effect of a variant on expression/accessibility, (2) Generating predicted coverage tracks for a locus, (3) Prioritising non-coding variants by predicted track delta. version: 1.0.0 type: worker license: Apache-2.0 category: biomodels requirements: [gpu] metadata: # SKILL.md loads johahi/borzoi-replicate-0 — a PyTorch port of Calico's # Borzoi ("ported weights (with permission)"). The HuggingFace model card # for that exact artifact states License: cc-by-4.0. Calico's CODE repo is # Apache-2.0, but the weights the skill downloads carry CC-BY-4.0. The model # card is where the license is declared (info_url — not a ToU page). # verified 2026-06-30 third_party:

• kind: weights

name: Borzoi (PyTorch port) provider: Calico Life Sciences license: CC-BY-4.0 info_url: https://huggingface.co/johahi/borzoi-replicate-0

Borzoi — DNA → Functional Track Prediction

Prerequisites

How to run

from borzoi_pytorch import Borzoi
model = Borzoi.from_pretrained("johahi/borzoi-replicate-0").cuda().eval()
# input: (batch, 4, 524288) one-hot DNA  → output: (batch, tracks, 6144) bins

Borzoi consumes ~524 kb one-hot windows and emits binned predictions across 7,611 human tracks (the separate 2,608-track mouse head is off by default; enable via enable_mouse_head=True and select with forward(..., is_human=False)). For variant scoring, run ref/alt windows centred on the variant and compare per-track output.

Output format

(B, T, L) tensor — T tracks × L 32-bp bins. Track metadata (assay, biosample) is in borzoi_pytorch.pytorch_borzoi_model.TRACKS_DF (or model.tracks_df when using the AnnotatedBorzoi subclass) — the base Borzoi model has no targets attribute.

Remote compute

Needs ≥24 GB VRAM and either pre-cached HF weights or egress to huggingface.co. Read compute_details({provider, mode:'read'}) for an environment with borzoi-pytorch, then:

c = host.compute.create(provider)
job = c.submit_job(
    intent="Borzoi track prediction for 1 locus — 1×GPU, ~2 min",
    inputs=[{"src": "borzoi_run.py", "dst_filename": "borzoi_run.py"}],
    command="python3 borzoi_run.py",   # env selection is host-specific — see compute_details for your provider
    outputs=["tracks.npz"],
    timeout_seconds=1800,
)
print(job.job_id)   # cell ends here — kernel never blocks on compute

Then call the wait_for_notification brain-tool. When the compute_done notification arrives, act on its payload:

save_artifacts(payload["featured_files"])   # paths under hpc/<job_id>/

For the full result dict (output_files, remote_workspace_root, …), re-enter the kernel: c.attach_job(job_id).result() then c.close(). See the remote-compute-ssh / remote-compute-modal skill for the orchestration details.

If the provider exposes a weight-cache mount, point HF_HOME at it inside borzoi_run.py (path is in compute_details).

Troubleshooting

Next: combine track deltas with evo2 likelihood deltas for a two-axis variant prioritisation.