ChromDragoNN

ChromDragoNN predicts genome-wide chromatin accessibility across diverse cellular contexts by integrating cis-regulatory DNA sequences and context-specific trans-regulator gene expression using multi-modal residual convolutional neural networks.

Key Features:

• Multi-modal residual neural network architecture: Employs a multi-modal residual convolutional neural network that integrates cis-regulatory DNA sequences and context-specific trans-regulator expression profiles for accessibility prediction.
• Generalization across cell types: Utilizes the average accessibility of genomic regions across training contexts as an explicit predictor to generalize predictions across diverse cell types.
• Enhanced prediction strategies: Implements novel model-training strategies to improve genome-wide prediction of both shared and context-specific chromatin accessible sites.
• Interpretability: Provides interpretability to identify cis- and trans-regulatory elements influencing chromatin accessibility across 123 diverse cellular contexts.

Scientific Applications:

• Genomics and epigenetics research: Enables genome-wide prediction of chromatin accessibility to support studies in genomics and epigenetics.
• Transcriptional regulation analysis: Facilitates investigation of transcriptional regulation and the roles of cis- and trans-regulatory elements.
• Cell-type regulatory inference: Supports analysis of gene expression variability and regulatory mechanisms across different cell types.

Methodology:

Trains multi-modal residual convolutional neural networks on genome-wide chromatin accessibility profiles and gene expression by integrating cis-regulatory DNA sequence inputs and context-specific trans-regulator expression, and incorporates the average accessibility of genomic regions across training contexts as a predictor.