Poly(A)-DG

Poly(A)-DG identifies polyadenylation signals (PAS) across species using a deep learning model with domain generalization to enable cross-species PAS prediction for studies of mRNA maturation and translation regulation.

Key Features:

• Deep Learning Architecture: Employs a Convolutional Neural Network–Multilayer Perceptron (CNN-MLP) to capture complex sequence patterns in PAS.
• Domain Generalization: Incorporates domain generalization techniques to generalize learned PAS patterns from training species to untrained target species without requiring retraining.
• Robustness to Limited Data: Maintains relatively high accuracy under insufficient or imbalanced dataset conditions and outperforms existing state-of-the-art methods.
• Cross-Species Validation: Validated using cross-species training sets, including experiments where two species were used for training and remaining species served as test cases.

Scientific Applications:

• Comparative Genomics: Enables analysis of evolutionary conservation and divergence of PAS sequences among different organisms.
• Functional Genomics: Supports studies of PAS roles in gene expression regulation, mRNA processing, and translation regulation.
• Analysis of Species with Limited Data: Extends PAS identification to species lacking extensive experimental PAS annotations by transferring models trained on other species.

Methodology:

Train a CNN-MLP network on annotated PAS data from selected species and apply a domain generalization technique to predict PAS in other species without additional training; validation used cross-species experiments (training on two species, testing on others) and evaluations under sufficient and limited data scenarios.