Research map / Probabilistic and scientific modeling / Molecules and proteins
LDARNet: DNA Adaptive Representation Network with Learnable Tokenization for Genomic Modeling
Daria Ledneva, Denis Kuznetsov
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 learned routing instead of fixed boundaries every 4 nucleotides (ablation) | LDARNet proxy 2.5M · NT H3K4me1 | MCC: +14.3pp on H3K4me1 (0.357 -> 0.500) | Improved |
| use learned routing instead of fixed boundaries (ablation) | LDARNet proxy 2.5M · NT H3K36me3 | MCC: +6.7pp (0.493 -> 0.560) | Improved |
| use learned routing instead of fixed boundaries (ablation) | LDARNet proxy 2.5M · NT H3 | MCC: +4.8pp (0.723 -> 0.771) | Improved |
| use learned routing instead of fixed boundaries (ablation) | LDARNet proxy 2.5M · NT Splice | MCC: 0.961 -> 0.901 | Hurt |
| use learned routing instead of fixed boundaries (ablation) | LDARNet proxy 2.5M · NT Promoter | MCC: 0.917 -> 0.916 | No clear effect |
| remove compression (nochunk, 4x FLOPs) versus learned compression N=4 (ablation) | LDARNet proxy 2.5M · NT H3K4me1 and H3K36me3 | MCC: +0.029 on H3K4me1, +0.037 on H3K36me3 | Mixed |
| add local attention layer to encoder (mt-M-m vs m-M-m) (ablation) | LDARNet proxy 2.5M · NT H3K4me1 | MCC: 0.448 -> 0.500 | Improved |
| add local attention layer to encoder (mt-M-m vs m-M-m) (ablation) | LDARNet proxy 2.5M · NT H3K36me3 | MCC: 0.521 -> 0.560 | Improved |
| use Transformer backbone instead of BiMamba-2 (mt-T-m vs mt-M-m) (ablation) | LDARNet proxy 2.5M · NT Splice | MCC: 0.901 -> 0.964 | Improved |
| use Transformer backbone instead of BiMamba-2 (mt-T-m vs mt-M-m) (ablation) | LDARNet proxy 2.5M · NT H3 histone | MCC: 0.771 -> 0.750 | Hurt |
| use attention-only outer modules (t-T-t) instead of BiMamba-2 (ablation) | LDARNet proxy 2.5M · NT H3K4me1 | MCC: 0.500 -> 0.434 | Hurt |
| lower compression ratio N=2 instead of N=4 (ablation) | LDARNet proxy 2.5M · NT H3K36me3 | MCC: 0.560 -> 0.598 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DNABERT-2 — DNABERT-2: Efficient Foundation Model and Benchmark For Multi-Species Genomes (ICLR 2024) · better on 2 of 2 shared results
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 0.87)
- DNABERT-2: Efficient Foundation Model and Benchmark For Multi-Species Genomes (ICLR 2024 · similarity 0.86)
- BEND: Benchmarking DNA Language Models on Biologically Meaningful Tasks (ICLR 2024 · similarity 0.86)
- JanusDNA: A Powerful Bi-directional Hybrid DNA Foundation Model (NeurIPS 2025 · similarity 0.86)
- LEMON-ZEST: Evolution-Informed Tokenization for Efficient Protein Language Modeling (NeurIPS 2026 · similarity 0.86)
- VQDNA: Unleashing the Power of Vector Quantization for Multi-Species Genomic Sequence Modeling (ICML 2024 · 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.
