Benchmark_Dataset-Primate_mammal_species_classification

## Summary This dataset provides a benchmark for evaluating the model's capability to identify species-specific genomic signatures across varying evolutionary distances. Using whole-genome assembly data from seven diverse mammals—including four primates (Human, Pan troglodytes, Pongo abelii, Callithrix) and three non-primates (Mouse, Cattle, Sheep)—we designed a multi-class species classification task. To rigorously assess generalization and prevent data leakage from homologous regions, we assigned distinct sets of chromosomes to the training, validation, and testing splits respectively. We sampled non-overlapping sequences across four fixed lengths—**1k, 8k, 32k, and 128k**—to ensure uniform genomic coverage. This task challenges the model to learn global phylogenetic features rather than memorizing specific genomic locations. ## Usage ```python from datasets import load_dataset #Download the whole dataset dataset = load_dataset("BGI-HangzhouAI/Benchmark_Dataset-Primate_mammal_species_classification") #Download a specific task task_name = "Primate_mammal_species_classification_1024" dataset = load_dataset( "BGI-HangzhouAI/Benchmark_Dataset-Primate_mammal_species_classification", data_files = { "train": f"{task_name}/train.jsonl", "test": f"{task_name}/test.jsonl", "eval": f"{task_name}/eval.jsonl", } ) ``` ## Detailed Information | Task | Task_name | Input fields | Train Seqs| Validation Seqs | Test Seqs | |:-------------:|:--------------:|:------------------:|:---------------:|:-------------------:|:--------------:| | Primate mammal species classification 1k | `Primate_mammal_species_classification_1024` | {seq, label} | 28,000 | 7,000 | 7,000 | | Primate mammal species classification 8k | `Primate_mammal_species_classification_8192` | {seq, label} | 28,000 | 7,000 | 7,000 | | Primate mammal species classification 32k | `Primate_mammal_species_classification_32768` | {seq, label} | 28,000 | 7,000 | 7,000 | | Primate mammal species classification 128k | `Primate_mammal_species_classification_131072` | {seq, label} | 28,000 | 7,000 | 7,000 | |Dataset Split | Chromosomes | |:-------------------:|:-----------------:| |Train |The rest of Autosomes| |Test |Chr 5, 7, 9, 10, 11 | |Validation |Chr 1, 3, 12, 13 | | Species | Label | Accession Numbers|Source| |:---------:|:-------------:|:----:|:-------:| |Human | 0 |[GWHCBHP00000000](https://ngdc.cncb.ac.cn/gwh/Assembly/37796/show)|CNCB| |Pan troglodytes| 1 |[GCA_028858775.3](https://www.ncbi.nlm.nih.gov/datasets/genome/GCA_028858775.3/)|NCBI| |Pongo abelii | 2 |[GCA_002880775.3](https://www.ncbi.nlm.nih.gov/datasets/genome/GCA_002880775.3/)|NCBI| |Callithrix | 3 |[GCA_049354665.1](https://www.ncbi.nlm.nih.gov/datasets/genome/GCA_049354665.1/)|NCBI| |Mouse | 4 |[GCA_964188535.1](https://www.ncbi.nlm.nih.gov/datasets/genome/GCA_964188535.1/)|NCBI| |Cattle | 5 |[GCF_002263795.3](https://www.ncbi.nlm.nih.gov/datasets/genome/GCF_002263795.3/)|NCBI| |Sheep | 6 |[GCF_016772045.2](https://www.ncbi.nlm.nih.gov/datasets/genome/GCF_016772045.2/)|NCBI| ## Data processing ### 1. Data Generation The dataset was constructed by partitioning genomes at the chromosome level to prevent data leakage. - Method: Sequences were sampled using a non-overlapping sliding window at fixed lengths **(1k, 8k, 32k, and 128k)**. - Quantity: **4,000** sequences for training per species. **1,000** sequences each for validation and testing per species. ### 2. Final format Datasets are saved in JSONL format. Each record contains the following fields: - "seq" — The DNA sequence string (A/C/G/T, uppercase). - "label" — A 7-class species indicator (0-6), representing the corresponding species. - "chr" — The source chromosome. - "len" — The fixed length of the sequence (e.g., 1024, 8192). - "species_name" — The name of the species associated with the label.