keras_dna: a wrapper for fast implementation of deep learning models in genomics

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning

Background:

• Deep learning is increasingly used for genomic annotation from DNA sequences.
• Challenges exist in preparing data for specific user-defined deep learning models.
• Existing tools may lack flexibility for diverse genomic data types and model architectures.

Purpose of the Study:

• To introduce keras_dna, a Python package designed for simplified Keras model implementation in genomics.
• To facilitate the handling of various standard bioinformatics file formats for deep learning model training.
• To support the development of novel genomic deep learning models with flexible input and output structures.

Main Methods:

• keras_dna processes standard bioinformatics file formats (bigwig, gff, bed, wig, bedGraph, fasta).
• It standardizes input data for Keras (TensorFlow high-level API) model training.
• The package supports the creation of models with single or multiple targets and inputs.

Main Results:

• Provides a user-friendly interface for building and training genomic deep learning models.
• Enables seamless integration of diverse genomic data types.
• Facilitates the development of both existing and new deep learning architectures for genomic annotation.

Conclusions:

• keras_dna addresses the need for accessible deep learning tools in genomics.
• It simplifies the workflow for researchers to build custom predictive models from DNA sequences.
• The package promotes further innovation in applying machine learning to genomic data analysis.