CNV-Finder Dataset

Example data for the CNV-Finder pipeline — an LSTM-based tool for large-scale identification of copy number variants (CNVs) from SNP array data.

Dataset Contents

SNP Metrics (snp_metrics/)

Per-sample signal intensity files containing Log R Ratio (LRR) and B Allele Frequency (BAF) extracted from genotyping arrays. Files are Hive-partitioned Parquet format organized by barcode, sample, and chromosome.

snp_metrics/
├── {barcode}/
│   └── {barcode}_{sample}/
│       ├── chromosome=1/
│       │   └── *.parquet
│       ├── chromosome=2/
│       ├── ...
│       ├── chromosome=22/
│       ├── chromosome=X/
│       ├── chromosome=Y/
│       └── chromosome=M/

Includes 20 samples across 2 cohorts (TEST1: barcode 2231, TEST2: barcode 4784), with 25 chromosomes each.

NBA Metadata (NBA_metadata/)

Reference metadata files containing repeating per-SNP values (SNP ID, position, GenTrain score) partitioned by chromosome. Used during the data preparation step of the pipeline.

NBA_metadata/
├── CHROM=1/
│   └── part.0.parquet
├── CHROM=2/
├── ...
└── CHROM=25/

Usage

Quick Download (Python)

from huggingface_hub import snapshot_download

# Download SNP metrics into example_data/
snapshot_download(
    repo_id="nkuznet/CNV-Finder",
    repo_type="dataset",
    allow_patterns="snp_metrics/**",
    local_dir="example_data"
)

# Download NBA metadata into ref_files/
snapshot_download(
    repo_id="nkuznet/CNV-Finder",
    repo_type="dataset",
    allow_patterns="NBA_metadata/**",
    local_dir="ref_files"
)

See run_pipeline.ipynb in the main repository for a full walkthrough.

Related

• GitHub Repository: nvk23/CNV-Finder
• SNP Metrics Generation: nvk23/SNP_metrics