DeepGeSeq: Deep learning library for Genomic Sequence modeling and analysis

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Deep learning shows promise in genomics for tasks like sequence activity prediction and variant effect quantification.
• Widespread adoption of deep learning in genomics is limited by complex model construction, training, and interpretation.
• DeepGeSeq is introduced as a user-friendly library for genomic sequence modeling and analysis.

Purpose of the Study:

• To introduce DeepGeSeq, a streamlined deep learning library for genomic sequence modeling.
• To reduce the computational learning curve associated with deep learning in genomics.
• To facilitate broader application of deep learning methods in biological discovery.

Main Methods:

• DeepGeSeq integrates state-of-the-art deep learning architectural modules.
• The library requires minimal user input via a configuration file and an intuitive agentic skill.
• Validation involved synthetic datasets, reproduction of existing models, and fine-tuning on user data.

Main Results:

• DeepGeSeq streamlines the deep learning workflow for genomic sequence analysis.
• Case studies demonstrated pipeline verification, model reproduction, and biological interpretation.
• The library showed versatility in single-cell ATAC-seq and MPRA data analysis, including cis-regulatory element dissection.

Conclusions:

• DeepGeSeq bridges the gap between computational complexity and biological discovery in genomics.
• The library provides an accessible resource for developing and applying deep learning in genomic research.
• DeepGeSeq empowers researchers to leverage advanced computational tools for genomic insights.