Research map / Architectures and efficiency / CNN design
Neural Edit Operations for Biological Sequences
Satoshi Koide, Keisuke Kawano, Takuro Kutsuna
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 |
|---|---|---|---|
| replace convolution inner product with differentiable Needleman-Wunsch score (EINN), gap cost 2.5 (ablation) | Tiny-CNN · CB513 (2% train data) | accuracy (%): 42.0 -> 43.0 | Improved |
| replace convolutions in first ConvBlock with EINN (ablation) | 2-block CNN with augmentation · CB513 | accuracy (%): 69.7 -> 69.8 | Improved |
| set EINN gap cost large (g>10), prohibiting insertion/deletion (ablation) | Tiny-EINN · CB513 (2% train data) | accuracy (%): 43.0 -> 42.0 | Hurt |
| increase depth from 2 to 12 blocks (ablation) | CNN with concatenation · CB513 | accuracy (%): 69.8 -> 71.5 | Improved |
| increase depth from 12 to 16 blocks (ablation) | CNN with concatenation · CB513 | accuracy (%): 71.5 -> 71.3 | No clear effect |
| use modified ConvBlock with concatenation (ablation) | 12-block CNN · CB513 | accuracy (%): 71.5 -> 71.5 | No clear effect |
| use residual (additive skip) blocks instead of concatenation (ablation) | deep CNN · CB513 | accuracy (%): 71.5 -> 71.0 | Hurt |
| random amino acid replacement data augmentation (15% of residues) (ablation) | deep CNN · CB513 | accuracy (%): up to 0.8-points | Improved |
| reduce training data to 1% (EINN vs CNN gain) (ablation) | Tiny-EINN · CB513 | accuracy gain over CNN: 1.4% at 1%, 1.0% at 2%, 0.6% at 5% | Improved |
| add pooling layers (ablation) | deep CNN · CB513 | accuracy (%) | No clear effect |
Closest papers by meaning
- GFlowNet Assisted Biological Sequence Editing (NeurIPS 2024 · similarity 0.85)
- Steering Protein Language Models (ICML 2025 · similarity 0.85)
- Revisiting Convolution Architecture in the Realm of DNA Foundation Models (ICLR 2025 · similarity 0.84)
- Neural Embeddings for kNN Search in Biological Sequence (AAAI 2024 · similarity 0.83)
- LEMON-ZEST: Evolution-Informed Tokenization for Efficient Protein Language Modeling (NeurIPS 2026 · similarity 0.83)
- Training Compute-Optimal Protein Language Models (NeurIPS 2024 · similarity 0.83)
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.
