SemanticCS

SemanticCS predicts transcription factor binding specificities from ChIP-seq datasets using deep learning to learn DNA sequence representations relevant to TF–DNA interactions.

Key Features:

• TF binding prediction: Predicts transcription factor–DNA binding probability from DNA sequences using a deep learning architecture.
• Training data: Trains on an ensemble of ChIP-seq data across multiple TFs and cell types using Multi-TF-cell datasets.
• Learned representations: Generates intermediate feature vectors of DNA sequences that capture regulatory information pertinent to transcription factor binding.
• Visualization interpretation: Interprets learned feature vectors through visualization analysis to aid understanding of underlying biological mechanisms.
• Performance benchmarking: Demonstrated superior performance compared to other popular methods in predicting TF binding specificities, validated by diverse experimental data and evaluation metrics.
• Variant effect evaluation: Evaluates how DNA sequence substitutions affect transcription factor binding affinity, exemplified with the RXR transcription factor.
• Representation reuse: Enables training of shallow machine learning models using its learned representations for downstream tasks.

Scientific Applications:

• Understanding TF–DNA interactions: Provides representations and predictions to study transcription factor binding specificities and mechanisms.
• Assessing regulatory variation: Identifies DNA substitutions that may lead to regulatory abnormalities by evaluating their effects on TF binding affinity, including RXR.
• Downstream genomic prediction: Supplies learned representations for training shallow machine learning models for related genomic studies and prediction tasks.

Methodology:

Applies deep learning architectures trained on ensemble Multi-TF-cell ChIP-seq datasets to learn intermediate feature vectors, uses visualization analysis to interpret these vectors, and employs the learned representations to train shallow machine learning models and to evaluate effects of DNA sequence substitutions on TF binding.