Genomic Interval Machine Learning

Acerca de

Geniml provides a specialized framework for applying unsupervised machine learning methods to genomic BED files, enabling researchers to perform similarity searches, clustering, and dimensionality reduction. It facilitates advanced workflows such as training Region2Vec models for genomic region embeddings, joint embedding of regions and metadata with BEDspace, and cell-level analysis for single-cell ATAC-seq data. By standardizing the creation of 'universes' (consensus peak sets) and providing robust evaluation utilities, it bridges the gap between raw genomic coordinates and high-dimensional vector representations for downstream analysis.

Características Principales

• scEmbed for single-cell ATAC-seq analysis and cell-type clustering
• BEDspace for joint embedding of genomic regions and associated metadata labels
• Comprehensive utilities for BED file randomization, caching, and embedding evaluation
• 2 GitHub stars
• Region2Vec for unsupervised genomic region embedding and dimensionality reduction
• Statistically rigorous consensus peak set (universe) construction methods

Casos de Uso

• Analyzing single-cell ATAC-seq datasets for cell-type annotation and clustering within scanpy workflows
• Performing dimensionality reduction and similarity analysis on large collections of BED files
• Creating searchable genomic databases by mapping metadata labels to region-set embeddings