Inference of the human polyadenylation code.

Area of Science:

• Computational Biology
• Genomics
• Molecular Biology

Background:

• Transcript processing involves cleavage at polyadenylation sites and addition of a poly(A)-tail.
• Alternative polyadenylation allows genes to produce transcript isoforms with varying 3'-ends.
• Accurate prediction of polyadenylation patterns is crucial for disease mutation identification and understanding regulatory mechanisms.

Purpose of the Study:

• To develop a computational model for predicting polyadenylation patterns from genomic features.
• To facilitate the identification and treatment of disease-causing mutations affecting polyadenylation.
• To understand sequence determinants underlying alternative polyadenylation.

Main Methods:

• Developed a deep learning model trained on 3'-end sequencing data.
• The model predicts tissue-specific polyadenylation site strength using only genomic sequence.
• Utilized competitive site selection data for model training.

Main Results:

• The deep learning model accurately predicts polyadenylation site selection in genes with multiple sites.
• The model successfully identifies candidate polyadenylation sites within the 3' untranslated region.
• Demonstrated utility in classifying variant pathogenicity and predicting antisense oligonucleotide effects.

Conclusions:

• The developed deep learning model offers a powerful tool for analyzing and predicting polyadenylation events.
• The model aids in understanding gene regulation and identifying disease-associated genetic variations.
• Provides insights into genomic regions critical for polyadenylation regulation at single-base resolution.