Nhoodie/hgt-real-mutation-pairs
收藏资源简介:
--- license: mit tags: - dna - mutation - hgt - cross-domain size_categories: - 1K<n<10K --- --- ## ℹ️ Dataset Note This is the primary ground-truth dataset for HGT mutation prediction. It contains real biological mutation pairs with authentic HGT signatures (Ti/Tv ratio ~0.77, dominated by C>T and T>C transitions consistent with viral polymerase error patterns). This dataset is **not** affected by the ICI-DC degeneracy issues identified in synthetic datasets. --- # HGT Real Mutation Pairs Real mutation pairs across 8 taxonomic domains for cross-domain horizontal gene transfer (HGT) mutation prediction. ## Dataset Details | Split | Pairs | Mean mutation rate | Mean distance | |-------|-------|-------------------|---------------| | train | 3,317 | ~4.2% | ~6 nt | | test | 826 | ~1.2% | 2.2 nt | ## Domains | Domain | Sequences | Rationale | |--------|-----------|-----------| | Archaea | 1,260 | Thermophile/halophile HGT | | Bacteria | 630 | Primary HGT domain | | Lichenes_HGT | 210 | Inter-kingdom HGT (fungi↔algae) | | Metazoa | 1,680 | Microbiome HGT | | Satellite_Viruses | 84 | Subviral agents | | Viridiplantae | 420 | Limited bacterial→plant HGT | | Viroids | 294 | Small RNA pathogens | | Viruses | 420 | Phage transduction | ## Format Each line is a JSON object: ```json {"parent": "ATGCGT...", "child": "ATGCGA...", "domain": "Bacteria", "distance": 3, "source_ids": ["AB123456"]} ``` ## Pipeline 1. Sequences fetched from NCBI with HGT-weighted sampling across 8 domains 2. Stop-codon-aware greedy chunking (~500bp) 3. Domain-aware nearest-neighbor pair building (C Levenshtein) 4. Leakage-free train/test split via connected-component clustering ## Related Datasets - [Nhoodie/hgt-bootstrap-v1-synthetic](https://huggingface.co/datasets/Nhoodie/hgt-bootstrap-v1-synthetic) — 8,112 synthetic pairs (ICI-DC v1) - [Nhoodie/hgt-bootstrap-v2-synthetic](https://huggingface.co/datasets/Nhoodie/hgt-bootstrap-v2-synthetic) — 256K synthetic pairs (ICI-DC v2)




