{"slug": "lc-seplm-long-range-contact-supervised-adaptation-for-sequence-only-protein", "title": "LC-SEPLM: long-range contact-supervised adaptation for sequence-only protein representation learning", "summary": "Researchers introduced LC-SEPLM (Long-range Contact-supervised ESM Protein Language Model), which adapts ESM2 with LoRA and long-range residue-pair contact supervision while retaining sequence-only downstream inference. Trained on 500,000 AlphaFold Swiss-Prot proteins, LC-SEPLM improved all eight protein-level tasks relative to ESM2, with the largest gain in remote-homology recognition where macro-F1 increased from 0.6122 to 0.6769. On the official ESM-S EC benchmark, LC-SEPLM outperformed ESM-S with a maximum absolute gain of 0.1771.", "body_md": "arXiv:2607.22777v1 Announce Type: new\nAbstract: Protein language models learn transferable sequence representations. However, because they primarily model contextual dependencies along amino-acid sequences, their training objectives do not explicitly constrain the model to learn three-dimensional residue contacts formed after folding . Here, we introduce LC-SEPLM (Long-range Contact-supervised ESM Protein Language Model), which adapts ESM2 with LoRA and long-range residue-pair contact supervision while retaining sequence-only downstream inference. Pair-specific queries use cross-attention over the complete sequence to extract global sequence context associated with long-range spatial contacts. To expose the model to diverse structural information, we trained LC-SEPLM on 500,000 AlphaFold Swiss-Prot proteins. In downstream evaluation, LC-SEPLM improved all eight protein-level tasks relative to ESM2. The largest gain occurred in remote-homology recognition, where macro-F1 increased from 0.6122 to 0.6769 (+0.0647, or 6.47 percentage points). On the official ESM-S EC benchmark, LC-SEPLM also outperformed ESM-S with a maximum absolute gain of 0.1771. These results support residue-pair contact supervision as a bounded route for introducing structural information into protein sequence representations while preserving sequence-only inference.", "url": "https://wpnews.pro/news/lc-seplm-long-range-contact-supervised-adaptation-for-sequence-only-protein", "canonical_source": "https://arxiv.org/abs/2607.22777", "published_at": "2026-07-28 04:00:00+00:00", "updated_at": "2026-07-28 04:13:02.939617+00:00", "lang": "en", "topics": ["machine-learning", "large-language-models", "artificial-intelligence"], "entities": ["LC-SEPLM", "ESM2", "LoRA", "AlphaFold", "Swiss-Prot", "ESM-S"], "alternates": {"html": "https://wpnews.pro/news/lc-seplm-long-range-contact-supervised-adaptation-for-sequence-only-protein", "markdown": "https://wpnews.pro/news/lc-seplm-long-range-contact-supervised-adaptation-for-sequence-only-protein.md", "text": "https://wpnews.pro/news/lc-seplm-long-range-contact-supervised-adaptation-for-sequence-only-protein.txt", "jsonld": "https://wpnews.pro/news/lc-seplm-long-range-contact-supervised-adaptation-for-sequence-only-protein.jsonld"}}