keras_dna

keras_dna provides conversion and data-generation functions to integrate standard bioinformatics file formats with the TensorFlow Keras high-level API for deep learning on genomic data.

Key Features:

• TensorFlow Keras integration: Interfaces with the TensorFlow Keras high-level API to implement and train deep learning models.
• Support for standard formats: Natively reads bigwig, gff, bed, wig, bedGraph, and fasta file formats.
• Flexible model inputs and targets: Supports models with single or multiple inputs and single or multiple targets for custom architectures.
• Keras-compatible data generators: Creates generators and data streams compatible with Keras models to optimize memory usage during training.
• On-the-fly conversion: Converts input formats on-the-fly to minimize the need for large volumes of stored intermediate data.

Scientific Applications:

• Genomic annotation prediction: Enables training of deep learning models for genomic annotation prediction from sequence and signal inputs.
• Gene expression analysis: Supports development of models for gene expression prediction and analysis.
• Variant calling: Can be employed in workflows using deep learning for variant calling.
• Functional genomics: Applicable to functional genomics studies linking sequence or signal features to genomic function.

Methodology:

Reads bigwig, gff, bed, wig, bedGraph, and fasta files and converts them into standardized inputs and Keras-compatible data generators/streams for training via the TensorFlow Keras high-level API.