alterlab-geniml

To generate genomic embeddings from BED files, you need a prepared universe BED file for tokenization. Once tokenized, you can train unsupervised embeddings to reduce dimensionality and enable region-level clustering and similarity searches.

Region2Vec is an embedding technique that learns vector representations for genomic regions from interval data. It reduces dimensionality to enable region-level analyses, similarity queries, and downstream machine learning tasks across BED files.

Yes, you can embed scATAC-seq data using the scEmbed workflow. It generates cell embeddings from single-cell genomic interval data, enabling unsupervised clustering and cell annotation across scATAC-seq datasets.

Yes, a universe BED file is required as a technical prerequisite for tokenization. It acts as a consensus peak reference to standardize genomic intervals before training embeddings and evaluating metadata.

The BEDspace workflow jointly embeds genomic regions and metadata labels for metadata-aware searches. This approach enables similarity queries across both regions and associated labels within bulk and single-cell datasets.

Universe building constructs consensus peak sets from genomic interval data to standardize tokenization references. This process creates a unified BED file required for cross-project tokenization and reproducible embedding pipelines.