geniml

To generate genomic embeddings from BED files, you can use unsupervised learning methods like Region2Vec to transform interval data into region-level feature vectors for similarity search and clustering. This requires tokenizing BED files against a predefined universe.

scEmbed is used for single-cell chromatin accessibility embeddings in scATAC-seq analysis to generate cell-level representations. These embeddings enable downstream clustering and similarity retrieval directly from sparse single-cell genomic interval data.

Yes, you can integrate metadata into genomic interval embeddings using the BEDspace method. BEDspace creates joint region and metadata embeddings, enabling metadata-aware queries and downstream machine learning tasks on BED collections.

To standardize tokenization across multiple genomic datasets, construct a consensus universe BED file. Universe-building workflows standardize tokenization across datasets, improving region-level embedding quality and ensuring consistent feature vector generation.

No, geniml does not require external dependencies to process genomic intervals and generate embeddings. It provides a self-contained Python API and CLI interface supporting tokenization, caching, evaluation, and search utilities.

The best way to retrieve similar genomic regions from a BED collection is by training unsupervised region embeddings using Region2Vec. These learned feature vectors enable direct region-level similarity search and retrieval.