license: cc-by-nc-4.0
task_categories:
- tabular-classification
- token-classification
language:
- en
tags:
- synthetic-data
- genomics
- bioinformatics
- dna
- sequencing
size_categories:
- 100M<n<1B
FreeSyntheticGenomicReads200M
A free dataset of 200 million fully synthetic short-read DNA sequences, built for developers, researchers, and students who need realistic sequencing-style data at scale — for testing bioinformatics pipelines, sequence-alignment tools, read-processing systems, or teaching computational biology. No real genomes, organisms, or individuals are represented in this data; every sequence is randomly generated.
Schema
| Column | Type | Description |
|---|---|---|
| read_id | string | Unique read identifier |
| sequence | string | 150 bp read sequence (A/C/G/T) |
| quality_scores | string | Per-base Phred+33 quality string (150 chars) |
| chromosome | string | Mapped chromosome (chr1-chr22, chrX, chrY, chrM), or * if unmapped |
| position | int | 1-based reference position, or 0 if unmapped |
| strand | string | + or - for mapped reads, * if unmapped |
| mapping_quality | int | MAPQ score (0-60) |
| gc_content | float | GC fraction of the read (0-1) |
| is_mapped | bool | Whether the read is mapped |
| read_length | int | Read length in bases (150) |
Format
Single Parquet file, Snappy compression, ~44 GB, 200,000,000 rows.
Quick Start
Note: this dataset is ~44 GB and 200M rows. Do not load it all at once — stream it in batches or query it with a columnar engine.
duckdb (recommended)
import duckdb
duckdb.sql("SELECT * FROM 'genomic_200M.parquet' LIMIT 10").show()
pyarrow batches
import pyarrow.parquet as pq
pf = pq.ParquetFile("genomic_200M.parquet")
for batch in pf.iter_batches(batch_size=100000):
... # process each batch
datasets (streaming)
from datasets import load_dataset
ds = load_dataset("ziadatalabs/FreeSyntheticGenomicReads200M", streaming=True)
Notes
Sequences are 150 bp random A/C/G/T with per-base Phred+33 quality strings skewed toward high quality, mirroring the shape of real short-read output. Reads are mapped (~94%) across chromosomes weighted by real human chromosome sizes, with mapping quality, position, strand, and GC content populated accordingly; unmapped reads carry the standard * / 0 placeholders. All entirely synthetic — no real biological sequence is represented.
License & Usage
Released under CC BY-NC 4.0 — personal, research, and educational use permitted, attribution required, no commercial use.
Created by Zia Data Labs. Questions or feedback: zia.data.team@protonmail.com