Predicting splicing patterns from the transcription factor binding sites in the promoter with deep learning

Area of Science:

• Molecular Biology
• Genomics
• Bioinformatics

Background:

• Alternative splicing generates transcriptome plasticity and proteome diversity in metazoan cells.
• The regulatory link between promoter-bound transcription factors and downstream alternative splicing remains largely unknown.

Purpose of the Study:

• To computationally investigate the regulatory relationship between promoter-bound transcription factor binding sites (TFBSs) and alternative splicing patterns.
• To explore how alterations in TFBSs at promoters affect exon splicing efficiency.

Main Methods:

• Utilized ENCODE data including DNase I hypersensitive site sequencing and transcriptomes across fifteen human tissues.
• Developed computational approaches and machine learning models, including convolutional neural networks (CNNs), to analyze TFBSs and splicing patterns.
• Performed in silico substitutions analysis and empirical validation using K562 CTCFL shRNA knock-down data.

Main Results:

• Machine learning models showed potential in predicting splicing patterns from TFBS occupancies, though generalization to singleton genes across tissues was limited.
• CNN models trained on promoter TFBS changes accurately predicted shifts in splicing patterns.
• In silico analysis identified potential splicing regulators, and empirical validation confirmed CTCFL's significant role in splicing regulation.

Conclusions:

• Promoter-bound TFBSs play a potential role in regulating downstream alternative splicing.
• This study provides insights for discovering novel alternative splicing regulatory mechanisms.