Composable foundations for genomic data.
CompoSeq is a collaborative initiative among maintainers of open-source genomics and machine-learning libraries. We're building a shared, high-performance data substrate that makes genomic data operate natively within the modern analytics and ML ecosystems: Apache Arrow, DataFusion, Polars, DuckDB, PyTorch, and more.
The goal is to make genomic workflows declarative, streaming, out-of-core, and agent-ready.
⚠️ Early and evolving. CompoSeq is a young initiative with a clear vision and an active roadmap. Follow along, open issues, and get involved.
CompoSeq coordinates work across these maintained, adopted projects (each lives in its home organization):
- oxbow — a format bridge that translates conventional genomic file formats into Apache Arrow record batches for any Arrow-native engine (Polars, DataFusion, DuckDB).
- polars-bio — a high-performance Python DataFrame library for genomics: interval operations powered by Apache DataFusion, a SQL frontend, indexed parallel I/O, out-of-core streaming, and cloud-native reads from S3/GCS.
- tangermeme — core operations for using and analyzing genomic ML models: sequence manipulation, feature attribution, variant effect scoring, and DNA design.
- bpnet-lite — compact, state-of-the-art sequence-to-function models for predicting chromatin accessibility, TF binding, and transcription initiation from DNA sequence.
- 🌐 Website: composeq.dev
- 💬 Questions & ideas: open an issue on any of our projects, or email hello@composeq.dev
- 🗺️ Roadmap: published at composeq.dev (coming soon)
We are committed to a welcoming, harassment-free community. Please read our Code of Conduct.
