DeepAraPPI

DeepAraPPI predicts protein-protein interactions (PPIs) in Arabidopsis thaliana by integrating sequence, domain, and Gene Ontology data using deep learning and natural language processing encodings.

Key Features:

• Siamese Recurrent Convolutional Neural Network (RCNN): Utilizes word2vec encoding to capture sequence-based features for PPI prediction.
• Multiple Layer Perceptron (MLP) with Domain2vec: Uses Domain2vec encoding to model domain-specific information relevant to PPIs.
• Multiple Layer Perceptron (MLP) with GO2vec: Employs GO2vec encoding to incorporate Gene Ontology data into predictions.
• Ensemble Integration: Combines outputs from the three models using a logistic regression model.
• Training and Test Sample Filtering: Applies strict filtering strategies to compile high-quality positive and negative training and test samples.
• Cross-Species Predictive Ability: Demonstrates improved cross-species PPI prediction performance in rice (Oryza sativa) relative to traditional machine learning methods.

Scientific Applications:

• Arabidopsis interactome mapping: Predicts PPIs to complement and extend experimentally identified interactomes in Arabidopsis thaliana.
• Protein function and interaction analysis: Supports inference of protein functions and interaction relationships within cellular processes.
• Comparative genomics: Enables comparative PPI studies between Arabidopsis thaliana and other plant species such as rice (Oryza sativa).

Methodology:

Processes sequence information with a Siamese RCNN using word2vec encoding; processes domain information with an MLP using Domain2vec encoding; processes Gene Ontology data with an MLP using GO2vec encoding; integrates the three model outputs via logistic regression; and applies strict filtering to construct high-quality positive and negative training and test samples.