Research map / Large language models / Language, safety and interpretability
AntigenLM: Structure-Aware DNA Language Modeling for Influenza
Yue Pei, Xuebin Chi, Yu Kang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| pretrain on full genomes instead of randomly cropped windows (incomplete genome) (ablation) | AntigenLM · Next-month HA/NA forecasting | token-level perplexity: 3.55 -> 1.26 | Improved |
| pretrain on whole genome with intact segments instead of single segments (ablation) | AntigenLM · Next-month HA/NA forecasting | token-level perplexity: 4.42 -> 1.26 | Improved |
| add non-HA/NA segments to pretraining instead of antigen-only (ablation) | AntigenLM · Next-month HA/NA forecasting | token-level perplexity: 4.56 -> 1.26 | Improved |
| preserve segment integrity (segment-wise and antigen-only vs full genome) for mismatch (ablation) | AntigenLM · Next-month forecasting amino-acid mismatch | mean AA mismatch | Improved |
| use AntigenLM instead of beth-1 and WHO recommendation | next-season forecasting (Japan) · HA/NA amino-acid mismatch | mismatch reduction: over 70% vs WHO, 50% vs beth-1 | Improved |
| use AntigenLM instead of beth-1 on unseen US region | geographic generalization · US NA mismatch | mean AA mismatch | Mixed |
| use incomplete-genome or segment-wise pretraining for subtype classification instead of full genome (ablation) | subtype classification · Subtype classification | subtype confusion | Hurt |
| pretrain on full genome vs ablations on H7N9 next-season forecast (ablation) | AntigenLM · H7N9 next-season forecasting | AA mismatch | No clear effect |
Closest papers by meaning
- DART-Eval: A Comprehensive DNA Language Model Evaluation Benchmark on Regulatory DNA (NeurIPS 2024 · similarity 0.88)
- BEND: Benchmarking DNA Language Models on Biologically Meaningful Tasks (ICLR 2024 · similarity 0.85)
- LEMON-ZEST: Evolution-Informed Tokenization for Efficient Protein Language Modeling (NeurIPS 2026 · similarity 0.84)
- HELM: Hierarchical Encoding for mRNA Language Modeling (ICLR 2025 · similarity 0.84)
- Absorb & Escape: Overcoming Single Model Limitations in Generating Heterogeneous Genomic Sequences (NeurIPS 2024 · similarity 0.84)
- Hyperbolic Genome Embeddings (ICLR 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
