SENSDeep

SENSDeep predicts protein-protein interaction sites (PPISs) using an ensemble deep learning approach to improve site-level identification from protein sequence and structural features.

Key Features:

• Ensemble Learning Model: Integrates multiple deep learning architectures including Recurrent Neural Networks (RNN), Convolutional Neural Networks (CNN), GRUs2s, GRUs2satt (GRU sequence-to-sequence with attention), and a multilayer perceptron.
• Feature Integration: Incorporates two additional embedded features—secondary structure and protein sequence—alongside twelve existing features in the training dataset.
• Performance Metrics: Reports improvements in sensitivity, F1 score, Matthews correlation coefficient (MCC), and area under the precision-recall curve (AUPRC) of up to 63.5% (sensitivity), 19.3% (F1), 18.5% (MCC), and 11.4% (AUPRC), respectively.
• Data Efficiency: Achieves comparable performance with less data when trained with the additional embedded features.
• Sliding Window Optimization: Explores various sliding window sizes to determine the optimal configuration for PPIS prediction.
• Comparative Analysis: Evaluated against structure-based methods, with some structure-based approaches showing superior performance in certain contexts.

Scientific Applications:

• PPIS Prediction: Prediction of protein-protein interaction sites to support analysis of molecular interactions.
• Protein Function Analysis: Facilitates investigation of protein functional mechanisms via predicted interaction sites.
• Drug Discovery: Supports target identification and interaction-site-focused drug design by identifying potential interaction interfaces.
• Network Biology: Aids elucidation of complex biological networks by mapping interaction sites across proteins.

Methodology:

An ensemble deep learning model combining RNN, CNN, GRUs2s, GRUs2satt, and a multilayer perceptron is trained on a dataset containing twelve existing features plus embedded secondary structure and protein sequence information, with optimization of sliding window sizes.