LangMoDHS

LangMoDHS predicts DNase I hypersensitive sites (DHSs) in the mouse genome using a deep learning language model to identify cis-regulatory elements involved in gene regulation.

Key Features:

• Deep learning architecture: Integrates a convolutional neural network (CNN), bi-directional long short-term memory (Bi-LSTM), and a feed-forward attention mechanism to model DNA sequence features.
• Language model framework: Employs a deep learning language model approach specifically applied to genomic sequences for DHS prediction.
• Parallel stacking of CNN and Bi-LSTM: Stacks CNN and Bi-LSTM in parallel to generate complementary sequence representations.
• Empirical validation: Evaluated using 5-fold cross-validation and independent tests across 14 tissues and four developmental stages in mouse with comparisons to iDHS-Deep.
• Sequence motif analysis: Uses indices related to information entropy to explore and characterize sequence motifs within DHSs.

Scientific Applications:

• Cis-regulatory element mapping: Prediction of DHSs to identify putative cis-regulatory elements involved in gene regulation.
• Tissue and developmental comparisons: Analysis of DHS landscapes across multiple tissues and four developmental stages in mouse.
• Gene expression and epigenetics studies: Supporting investigations of gene regulation, gene expression analysis, and epigenetic modification patterns via DHS prediction.
• Motif characterization: Characterization of sequence motifs associated with DHSs using information entropy-based indices.

Methodology:

Convolutional neural network (CNN), bi-directional LSTM (Bi-LSTM), a feed-forward attention mechanism, parallel stacking of CNN and Bi-LSTM, evaluation by 5-fold cross-validation and independent tests across 14 tissues and four developmental stages, and sequence motif analysis using indices related to information entropy.