Research map / Probabilistic and scientific modeling / Molecules and proteins
MergeDNA: Context-Aware Genome Modeling with Dynamic Tokenization Through Token Merging
Siyuan Li, Kai Yu, Anna Wang, Zicheng Liu, Chang Yu, Jingbo Zhou, Qirong Yang, Yucheng Guo, Xiaoming Zhang, Stan Z. Li
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 first Transformer layers with the Local Encoder (local token merging) (ablation) | MergeDNA · Genomic Benchmarks (8 tasks avg) | accuracy gain: +0.39 | Improved |
| add Merged Token Reconstruction objectives and AMTM (ablation) | MergeDNA · Genomic Benchmarks (8 tasks avg) | accuracy gain: +1.03 | Improved |
| add Adaptive Masked Token Modeling on filtered tokens (ablation) | MergeDNA · Genomic Benchmarks (8 tasks avg) | accuracy gain: +1.03 -> +1.57 | Improved |
| down-weight latent MTR loss with lambda=0.25 (ablation) | MergeDNA · Genomic Benchmarks (8 tasks avg) | accuracy gain: +1.03 -> +1.57 | Improved |
| use 4 local merging layers instead of 2 (ablation) | MergeDNA · Genomic Benchmarks (8 tasks avg) | accuracy gain: +1.21 -> +1.57 | Improved |
| use MergeDNA instead of MxDNA | dynamic tokenizer · NT Benchmark (18 tasks avg) | MCC/F1: 78.14 -> 78.39 | Improved |
| use MergeDNA (380M) instead of HybriDNA-7B | DNA foundation model · GUE (24 tasks avg) | MCC/F1: 76.42 -> 77.11 | Improved |
| use MergeDNA instead of Evo2-7B | DNA foundation model · Protein fitness (human, zero-shot) | SRCC: 36.9 -> 20.58 | Hurt |
| use MergeDNA instead of GENERator | DNA foundation model · Genomic Benchmarks (8 tasks avg) | top-1 accuracy: 90.71 -> 90.87 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- DNABERT2 — DNABERT-2: Efficient Foundation Model and Benchmark For Multi-Species Genomes (ICLR 2024) · better on 2 of 2 shared results
- VQDNA — VQDNA: Unleashing the Power of Vector Quantization for Multi-Species Genomic Sequence Modeling (ICML 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Model Decides How to Tokenize: Adaptive DNA Sequence Tokenization with MxDNA (NeurIPS 2024 · similarity 0.90)
- DNABERT-2: Efficient Foundation Model and Benchmark For Multi-Species Genomes (ICLR 2024 · similarity 0.88)
- VQDNA: Unleashing the Power of Vector Quantization for Multi-Species Genomic Sequence Modeling (ICML 2024 · similarity 0.87)
- MrT5: Dynamic Token Merging for Efficient Byte-level Language Models (ICLR 2025 · similarity 0.86)
- GeneZip: Region-Aware Compression for Long Context DNA Modeling (NeurIPS 2026 · similarity 0.86)
- BEND: Benchmarking DNA Language Models on Biologically Meaningful Tasks (ICLR 2024 · similarity 0.85)
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.
