SemanticBI

SemanticBI predicts transcription factor (TF)-DNA binding intensities using deep learning to quantify relative binding preferences.

Key Features:

• Deep Learning Architecture: Employs a convolutional neural network (CNN) combined with a recurrent neural network (RNN) in a CNN-RNN hybrid to capture spatial and sequential patterns in DNA sequences.
• Training Data: Trained on an ensemble of protein binding microarray datasets from the DREAM5 challenge encompassing multiple transcription factors.
• Quantitative Binding Predictions: Produces quantitative predictions of TF-DNA binding intensities to estimate relative binding preferences.
• Vectorized Sequence-Oriented Features: Learns and utilizes vectorized sequence-oriented feature representations of DNA sequences that provide insights beyond traditional motif-based approaches.
• Predictive Performance: Demonstrates superior performance compared to existing methods in predicting TF-DNA binding intensities.

Scientific Applications:

• Transcriptional regulatory network analysis: Quantifies TF-DNA binding intensities to inform reconstruction and analysis of transcriptional regulatory networks.
• Gene expression regulation studies: Aids investigation of nuanced regulation of gene expression by providing quantitative TF binding information.
• Developmental biology: Supports studies of developmental processes by enabling analysis of TF binding dynamics relevant to development.
• Disease mechanism investigation: Facilitates exploration of disease mechanisms where altered TF binding influences gene regulation.
• Synthetic biology: Informs synthetic biology design by supplying quantitative TF-DNA binding preferences for regulatory element engineering.

Methodology:

Implements a CNN-RNN hybrid architecture trained on protein binding microarray datasets from the DREAM5 challenge and learns vectorized sequence-oriented features from DNA sequences.