Research map / Large language models / Instruction and fine-tuning
Genome-Factory: A Library for Tuning, Deploying, and Interpreting Genomic Foundation Models
Weimin Wu, Xuefeng Song, Yibo Wen, Qinjie Lin, Zhihan Zhou, Jerry Yao-Chieh Hu, Zhong Wang, Han Liu
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 adapter tuning instead of full fine-tuning (ablation) | DNABERT-2 · COVID variant (GUE) | peak memory and throughput: 7.30 GB -> 2.09 GB memory; 40.70 -> 124.30 KTok/s | Improved |
| use LoRA instead of full fine-tuning (ablation) | DNABERT-2 · COVID variant (GUE) | trainable params and throughput: 117.08M -> 1.49M params; 40.70 -> 44.07 KTok/s | Improved |
| scale model from Nucleotide Transformer-500M to 2.5B under LoRA (ablation) | LoRA · Efficiency vs full tuning | memory savings and throughput ratio: memory savings 16% -> 32%; throughput 1.20 -> 1.39 times | Improved |
| scale GenomeOcean from 100M to 500M under LoRA (ablation) | LoRA · Efficiency vs full tuning | memory savings and throughput ratio: memory savings 5% -> 21%; throughput 1.01 -> 1.27 times | Improved |
| use LoRA instead of full fine-tuning (ablation) | GenomeOcean-100M · Genomic Benchmarks | MCC: 68.26 -> 66.28 | No clear effect |
| use LoRA instead of full fine-tuning (ablation) | DNABERT-2 · GUE | GUE score: 65.25 -> 48.39 | Hurt |
| use adapter instead of LoRA (ablation) | Caduceus-131k · GUE | GUE score: 34.70 -> 38.61 | Improved |
| use adapter instead of LoRA (ablation) | GenomeOcean-100M · GUE | GUE score: 59.35 -> 46.65 | Hurt |
| use adapter tuning instead of full fine-tuning (ablation) | DNABERT-2, HyenaDNA, Nucleotide Transformer · GUE | memory, throughput, GUE score | Mixed |
| use foundation model (GenomeOcean-100M full) over classical SVM baseline | GUE · GUE | GUE score: 57.01 -> 65.52 | Improved |
| use larger EVO-1-131k (7B) with LoRA rather than smaller models | LoRA · GUE | GUE score: 44.97 vs 59.35 (GenomeOcean-100M) | Hurt |
| use GenomeOcean-4B instead of EVO 2-7B | LoRA on GUE epigenetic marks · GUE epigenetic marks (10 tasks) | average MCC: 61.77 -> 60.43 | No clear effect |
Closest papers by meaning
- 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 0.85)
- Fast and Low-Cost Genomic Foundation Models via Outlier Removal (ICML 2025 · similarity 0.84)
- Hyperbolic Genome Embeddings (ICLR 2025 · similarity 0.84)
- Training Compute-Optimal Protein Language Models (NeurIPS 2024 · similarity 0.83)
- Absorb & Escape: Overcoming Single Model Limitations in Generating Heterogeneous Genomic Sequences (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.
