Geniml is a Python package and Claude Code skill designed for building machine learning models on genomic interval data, primarily BED files, for tasks like clustering and embedding.

What are the core ML architectures used by Geniml?

Geniml utilizes several architectures, including word2vec-style learning for Region2Vec and StarSpace for joint embeddings in BEDspace.

What are 'universes' in the context of Geniml?

A 'universe' is a consensus peak set created from a collection of BED files. It acts as a reference for tokenization and standardizes regions across different genomic datasets.

Does Geniml support searching genomic data by metadata?

Yes, the BEDspace feature allows for the creation of joint embeddings for regions and metadata, enabling cross-modal queries like finding regions associated with specific tissue types.

Can I use Geniml for single-cell ATAC-seq analysis?

Yes, Geniml includes the scEmbed module, which is specifically designed to train cell-level embeddings for scATAC-seq data, integrating seamlessly with scanpy workflows.

Geniml: Genomic Machine Learning

Name: Geniml: Genomic Machine Learning
Author: Sologa

bySologa

数据科学与机器学习

Builds machine learning models and embeddings for genomic interval data to enable similarity searches, clustering, and single-cell analysis.

关于

Geniml is a specialized toolkit for performing unsupervised machine learning on genomic regions stored in BED files. It provides a suite of advanced capabilities including Region2Vec for genomic region embeddings, BEDspace for joint region-metadata representations, and scEmbed for single-cell ATAC-seq analysis. This skill enables researchers and developers to build consensus 'universes' (reference peak sets), perform dimensionality reduction on chromatin accessibility data, and execute cross-modal genomic queries, all within a streamlined, statistically rigorous framework designed for high-performance bioinformatics workflows.

主要功能

Region2Vec: Unsupervised genomic region embeddings for similarity and feature engineering.
Consensus Peaks: Statistical tools for building reference peak sets (universes) using HMM and ML methods.
Genomic Utilities: Tools for BED file caching (BBClient), randomization (BEDshift), and embedding evaluation.
0 GitHub stars
BEDspace: Jointly embed genomic regions and metadata labels for multi-modal searching.
scEmbed: Cell-level embeddings optimized for single-cell ATAC-seq and scanpy integration.

使用场景

Dimensionality reduction and clustering of single-cell ATAC-seq datasets for cell-type annotation.
Standardizing heterogeneous BED file collections into a unified consensus reference for downstream ML tasks.
Building searchable genomic databases that link experimental metadata to specific interval patterns.

关于

主要功能

Region2Vec: Unsupervised genomic region embeddings for similarity and feature engineering.
Consensus Peaks: Statistical tools for building reference peak sets (universes) using HMM and ML methods.
Genomic Utilities: Tools for BED file caching (BBClient), randomization (BEDshift), and embedding evaluation.
0 GitHub stars
BEDspace: Jointly embed genomic regions and metadata labels for multi-modal searching.
scEmbed: Cell-level embeddings optimized for single-cell ATAC-seq and scanpy integration.

使用场景

Dimensionality reduction and clustering of single-cell ATAC-seq datasets for cell-type annotation.
Standardizing heterogeneous BED file collections into a unified consensus reference for downstream ML tasks.
Building searchable genomic databases that link experimental metadata to specific interval patterns.