pith:65PALSYV
Protein Circuit Tracing via Cross-layer Transcoders
ProtoMech applies cross-layer transcoders to protein language models to recover 82-89% of model performance using sparse circuits that match biological motifs and improve protein design in over 70% of cases.
arxiv:2602.12026 v2 · 2026-02-12 · cs.LG · q-bio.QM
Add to your LaTeX paper
\usepackage{pith}
\pithnumber{65PALSYVPD7Q6GMTXKZI7PB2OY}
Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge
Record completeness
Claims
ProtoMech recovers 82-89% of the original performance on protein family classification and function prediction tasks. ProtoMech then identifies compressed circuits that use <1% of the latent space while retaining up to 79% of model accuracy... Steering along these circuits enables high-fitness protein design, surpassing baseline methods in more than 70% of cases.
That the sparse latent representations learned jointly across layers faithfully approximate the model's full computational circuitry and that the identified circuits correspond to genuine structural and functional motifs rather than artifacts of the transcoder training.
ProtoMech applies cross-layer transcoders to protein language models to recover 82-89% of model performance using sparse circuits that match biological motifs and improve protein design in over 70% of cases.
Formal links
Cited by
Receipt and verification
| First computed | 2026-05-18T03:09:23.608707Z |
|---|---|
| Builder | pith-number-builder-2026-05-17-v1 |
| Signature | Pith Ed25519
(pith-v1-2026-05) · public key |
| Schema | pith-number/v1.0 |
Canonical hash
f75e05cb1578ff0f1993bab28fbc3a7612518f740638b6a8b291df66c67a32bb
Aliases
· · · · ·Agent API
Verify this Pith Number yourself
curl -sH 'Accept: application/ld+json' https://pith.science/pith/65PALSYVPD7Q6GMTXKZI7PB2OY \
| jq -c '.canonical_record' \
| python3 -c "import sys,json,hashlib; b=json.dumps(json.loads(sys.stdin.read()), sort_keys=True, separators=(',',':'), ensure_ascii=False).encode(); print(hashlib.sha256(b).hexdigest())"
# expect: f75e05cb1578ff0f1993bab28fbc3a7612518f740638b6a8b291df66c67a32bb
Canonical record JSON
{
"metadata": {
"abstract_canon_sha256": "a2da0fa6107cab3e2935b382d332a6affcbd602a471fa13b37e10674a68fd474",
"cross_cats_sorted": [
"q-bio.QM"
],
"license": "http://creativecommons.org/licenses/by-nc-nd/4.0/",
"primary_cat": "cs.LG",
"submitted_at": "2026-02-12T14:57:57Z",
"title_canon_sha256": "8986ec24dc1c664190b4d340f60ae0240eaf4cebea73a7b7250faaa7de5c7dec"
},
"schema_version": "1.0",
"source": {
"id": "2602.12026",
"kind": "arxiv",
"version": 2
}
}