{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:JW24JOYSN4KKVMWAS63SY2SYPP","short_pith_number":"pith:JW24JOYS","schema_version":"1.0","canonical_sha256":"4db5c4bb126f14aab2c097b72c6a587bf94ee2d345091158920c9cdce7805487","source":{"kind":"arxiv","id":"2503.08764","version":1},"attestation_state":"computed","paper":{"title":"Towards Interpretable Protein Structure Prediction with Sparse Autoencoders","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"q-bio.BM","authors_text":"David J. Yang, John J. Yang, Nithin Parsan","submitted_at":"2025-03-11T17:57:29Z","abstract_excerpt":"Protein language models have revolutionized structure prediction, but their nonlinear nature obscures how sequence representations inform structure prediction. While sparse autoencoders (SAEs) offer a path to interpretability here by learning linear representations in high-dimensional space, their application has been limited to smaller protein language models unable to perform structure prediction. In this work, we make two key advances: (1) we scale SAEs to ESM2-3B, the base model for ESMFold, enabling mechanistic interpretability of protein structure prediction for the first time, and (2) w"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2503.08764","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"q-bio.BM","submitted_at":"2025-03-11T17:57:29Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"1d71d90763655e4d8c237bc8e91bdd095b0d7804b92d271660c4f0f1e9401edf","abstract_canon_sha256":"1e181656210ae0fd5fadbe736fd32ee356814941aece24945acaa61384c73767"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:29:39.988676Z","signature_b64":"J8uYVcbrm/3CKRt9kBIs9JYsOCGs5CI/8D1wlcIbpoVShI0SuKyPhSV7rkn+jO8DE/9pdPhfdlhpBTv7XcDhAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4db5c4bb126f14aab2c097b72c6a587bf94ee2d345091158920c9cdce7805487","last_reissued_at":"2026-07-05T10:29:39.988098Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:29:39.988098Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Towards Interpretable Protein Structure Prediction with Sparse Autoencoders","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"q-bio.BM","authors_text":"David J. Yang, John J. Yang, Nithin Parsan","submitted_at":"2025-03-11T17:57:29Z","abstract_excerpt":"Protein language models have revolutionized structure prediction, but their nonlinear nature obscures how sequence representations inform structure prediction. While sparse autoencoders (SAEs) offer a path to interpretability here by learning linear representations in high-dimensional space, their application has been limited to smaller protein language models unable to perform structure prediction. In this work, we make two key advances: (1) we scale SAEs to ESM2-3B, the base model for ESMFold, enabling mechanistic interpretability of protein structure prediction for the first time, and (2) w"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.08764","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2503.08764/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2503.08764","created_at":"2026-07-05T10:29:39.988169+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.08764v1","created_at":"2026-07-05T10:29:39.988169+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.08764","created_at":"2026-07-05T10:29:39.988169+00:00"},{"alias_kind":"pith_short_12","alias_value":"JW24JOYSN4KK","created_at":"2026-07-05T10:29:39.988169+00:00"},{"alias_kind":"pith_short_16","alias_value":"JW24JOYSN4KKVMWA","created_at":"2026-07-05T10:29:39.988169+00:00"},{"alias_kind":"pith_short_8","alias_value":"JW24JOYS","created_at":"2026-07-05T10:29:39.988169+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.12209","citing_title":"Interpretable enzyme function prediction via sparse autoencoder features of ESMC across the microbial protein universe","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2606.10080","citing_title":"VFUSE: Virulent Feature Understanding with Sparse autoEncoders","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16331","citing_title":"Retrieval and competition: how a protein foundation model starts a protein","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16331","citing_title":"Retrieval and competition: how a protein foundation model starts a protein","ref_index":10,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/JW24JOYSN4KKVMWAS63SY2SYPP","json":"https://pith.science/pith/JW24JOYSN4KKVMWAS63SY2SYPP.json","graph_json":"https://pith.science/api/pith-number/JW24JOYSN4KKVMWAS63SY2SYPP/graph.json","events_json":"https://pith.science/api/pith-number/JW24JOYSN4KKVMWAS63SY2SYPP/events.json","paper":"https://pith.science/paper/JW24JOYS"},"agent_actions":{"view_html":"https://pith.science/pith/JW24JOYSN4KKVMWAS63SY2SYPP","download_json":"https://pith.science/pith/JW24JOYSN4KKVMWAS63SY2SYPP.json","view_paper":"https://pith.science/paper/JW24JOYS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.08764&json=true","fetch_graph":"https://pith.science/api/pith-number/JW24JOYSN4KKVMWAS63SY2SYPP/graph.json","fetch_events":"https://pith.science/api/pith-number/JW24JOYSN4KKVMWAS63SY2SYPP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JW24JOYSN4KKVMWAS63SY2SYPP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JW24JOYSN4KKVMWAS63SY2SYPP/action/storage_attestation","attest_author":"https://pith.science/pith/JW24JOYSN4KKVMWAS63SY2SYPP/action/author_attestation","sign_citation":"https://pith.science/pith/JW24JOYSN4KKVMWAS63SY2SYPP/action/citation_signature","submit_replication":"https://pith.science/pith/JW24JOYSN4KKVMWAS63SY2SYPP/action/replication_record"}},"created_at":"2026-07-05T10:29:39.988169+00:00","updated_at":"2026-07-05T10:29:39.988169+00:00"}