DeeReCT-APA

DeeReCT-APA predicts quantitative usage levels of alternative polyadenylation sites (PAS) within genes to characterize APA-mediated post-transcriptional regulation affecting mRNA stability, localization, and translation.

Key Features:

• Quantitative Prediction: Outputs percentage scores representing the relative usage levels of all PAS within a gene as a regression target.
• Deep Learning Architecture: Implements a convolutional neural network (CNN) for sequence feature extraction combined with a bidirectional long short-term memory (LSTM) network to model interactions among competing PAS.
• Variable-Length Target: Handles genes with varying numbers of PAS by formulating APA prediction as a variable-length regression problem.
• Performance Superiority: Demonstrates superior accuracy over existing methods in pairwise comparison, highest-usage prediction, and ranking tasks.
• Genetic Variation Analysis: Predicts the impact of genetic variations on APA patterns to aid mechanistic interpretation of APA regulation.

Scientific Applications:

• Post-transcriptional regulation analysis: Quantitatively assesses PAS usage to study APA-mediated regulation of mRNA stability, localization, and translation.
• Gene expression and cellular function studies: Investigates how alternative PAS selection alters gene expression and contributes to cellular diversity.
• Variant interpretation and disease research: Interprets effects of genetic variants on APA to identify potential regulatory mechanisms relevant to biological processes and diseases.

Methodology:

Combines CNNs for sequence feature extraction and bidirectional LSTMs to model interactions among PAS, trained as a regression model with variable-length targets that outputs percentage usage scores for all PAS within a gene.