DeepPASTA

DeepPASTA predicts polyadenylation (polyA) sites in pre-mRNA by integrating sequence and RNA secondary structure information to identify tissue-specific and dominant polyA sites and characterize alternative polyadenylation affecting 3' UTRs.

Key Features:

• Integration of sequence and structure data: Combines nucleotide sequence features with RNA secondary structure information to improve polyA site prediction accuracy.
• Tissue-specific prediction: Produces predictions of polyA sites specific to tissue contexts to capture context-dependent alternative polyadenylation.
• Dominance prediction: Predicts the most dominant polyA site for a gene in a given tissue and assesses relative dominance between two polyA sites of the same gene.

Scientific Applications:

• Alternative polyadenylation analysis: Enables identification and characterization of alternative polyadenylation events across genes and tissues.
• 3' UTR length and disease association: Facilitates studies linking 3' UTR shortening or lengthening to disease mechanisms and gene regulation changes.
• Gene regulation and transcriptome studies: Supports investigation of how polyadenylation site choice affects mRNA stability, localization, translation, and binding interactions.

Methodology:

DeepPASTA employs a deep neural network that integrates sequence data and RNA secondary structure information, is trained on extensive datasets, and has been validated through experiments demonstrating superior performance compared to existing prediction tools.