Research map / Probabilistic and scientific modeling / Molecules and proteins
PatchDNA: A Flexible and Biologically-Informed Alternative to Tokenization for DNA
Alice Del Vecchio, Chantriolnt-Andreas Kapourani, Abdullah M Athar, Agnieszka Dobrowolska, Andrew Anighoro, Benjamin Tenmann, Lindsay Edwards, Cristian Regep
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 |
|---|---|---|---|
| use conservation-based patching instead of entropy-based patching (ablation) | PatchDNA · DART-Eval Task 2 | accuracy: 0.650 -> 0.725 | Improved |
| use conservation-based patching instead of entropy-based patching (ablation) | PatchDNA · BEND gene finding | MCC: 0.37 -> 0.58 | Improved |
| use conservation-based patching instead of fixed patch size 20 (ablation) | PatchDNA · NT promoter all (linear probe) | MCC: 0.719 -> 0.779 | Improved |
| use conservation-based patching instead of fixed patch size 20 (ablation) | PatchDNA · DART-Eval Task 3 | accuracy: 0.477 -> 0.457 | Hurt |
| use conservation-based patching instead of entropy patching (ablation) | PatchDNA · NT histone tasks (H2AFZ) | MCC: 0.401 -> 0.396 | No clear effect |
| re-patch to single nucleotide at fine-tuning for splice sites | PatchDNA · NT splice sites acceptors | MCC: 0.754 -> 0.946 | Improved |
| add DNase-aware re-patching at fine-tuning | PatchDNA-7M · Cell-type CAGE K562 | Pearson: 0.754 -> 0.828 | Improved |
| add DNase-aware re-patching at fine-tuning | PatchDNA-7M · Cell-type CAGE Hepatocyte | Pearson: 0.717 -> 0.727 | Improved |
| patch with mismatched cell-type DNase-seq | PatchDNA-7M · Cell-type CAGE K562 | Pearson: 0.828 -> 0.775 | Hurt |
| add cCRE-aware re-patching at fine-tuning | PatchDNA-7M · CAGE prediction | Pearson: cell 0.771 -> 0.792; full 0.471 -> 0.408 | Mixed |
| use patching with BLT architecture instead of tokenization | PatchDNA · DART-Eval | mean rank: best mean rank 2.0 | Improved |
Closest papers by meaning
- Model Decides How to Tokenize: Adaptive DNA Sequence Tokenization with MxDNA (NeurIPS 2024 · similarity 0.91)
- BEND: Benchmarking DNA Language Models on Biologically Meaningful Tasks (ICLR 2024 · similarity 0.88)
- DNABERT-2: Efficient Foundation Model and Benchmark For Multi-Species Genomes (ICLR 2024 · similarity 0.88)
- DART-Eval: A Comprehensive DNA Language Model Evaluation Benchmark on Regulatory DNA (NeurIPS 2024 · similarity 0.87)
- LEMON-ZEST: Evolution-Informed Tokenization for Efficient Protein Language Modeling (NeurIPS 2026 · similarity 0.86)
- JanusDNA: A Powerful Bi-directional Hybrid DNA Foundation Model (NeurIPS 2025 · similarity 0.86)
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.
