TRAPT: a multi-stage fused deep learning framework for predicting transcriptional regulators based on large-scale...

Area of Science:

• Genomics
• Computational Biology
• Epigenetics

Background:

• Identifying transcriptional regulators (TRs) is vital for understanding gene expression, disease mechanisms, and cellular processes.
• Context-specific TR identification remains challenging due to the complexity of regulatory elements and epigenomic signals.

Purpose of the Study:

• To develop and validate a novel deep learning framework, Transcription Regulator Activity Prediction Tool (TRAPT), for inferring TR activity from multi-omics epigenomic data.
• To enhance the prediction accuracy of TRs, including transcription co-factors and chromatin regulators.

Main Methods:

• TRAPT integrates regulatory potentials from cis-regulatory elements and genome-wide binding sites using a multi-modality deep learning approach.
• The framework was evaluated on 570 TR-related datasets, comparing its performance against state-of-the-art methods.

Main Results:

• TRAPT demonstrated superior performance in predicting TRs compared to existing methods.
• The tool showed particular strength in forecasting transcription co-factors and chromatin regulators.
• Key TRs associated with diseases, genetic variations, cell-fate decisions, and specific tissues were successfully identified.

Conclusions:

• TRAPT offers an innovative and effective approach for identifying transcriptional regulators by leveraging large-scale epigenomic data.
• This method provides valuable insights into the roles of TRs in various biological contexts, including disease pathogenesis.