D-miRT

D-miRT predicts condition-specific microRNA (miRNA) transcription start sites (TSSs) by integrating low-resolution epigenetic data (DNase-Seq and histone modification profiles) with high-resolution sequence features using a two-stream convolutional neural network.

Key Features:

• Two-stream convolutional neural network: Uses a dual-branch CNN architecture to learn separate representations from epigenetic and sequence inputs.
• Multi-resolution data integration: Combines low-resolution epigenetic signals (DNase-Seq, histone modification profiles) with high-resolution nucleotide sequence features for prediction.
• Condition-specific prediction: Produces TSS predictions that are specific to biological conditions or cell types represented in the input data.
• Benchmarking against alternatives: Evaluated against alternative computational models and reported superior performance on multiple training datasets.
• Generalization across cell lines: Demonstrated ability to identify cell-specific miRNA TSSs in cell lines not included in model training.

Scientific Applications:

• Condition-specific miRNA TSS identification: Locates transcription start sites of miRNAs under specific biological conditions or cell types.
• Annotation of distal miRNA regulatory elements: Supports annotation of miRNA promoters that can be located tens of thousands of nucleotides from precursor miRNAs.
• Cross-cell-type comparative studies: Enables comparison of miRNA TSS usage across diverse biological conditions and cell lines to study gene regulation.

Methodology:

Implements a two-stream convolutional neural network that integrates DNase-Seq and histone modification profiles with high-resolution sequence features, and is trained and benchmarked against alternative computational models with evaluations including cell lines excluded from training.