DeepTP

DeepTP predicts thermophilic proteins from protein sequences using deep learning to identify thermostable proteins for enzyme engineering and biopharmaceutical applications.

Key Features:

• Self-Attention Mechanism: Incorporates a self-attention mechanism to assign different weights to features extracted from protein sequences.
• Multiple-Channel Feature Fusion: Employs multiple-channel feature fusion to integrate various biological features for comprehensive protein characterization.
• Convolutional and Recurrent Neural Networks: Utilizes convolutional neural networks (CNNs) and bidirectional long short-term memory networks (BiLSTMs) to extract spatial and sequential dependencies from sequences.
• Training Dataset: Trained on a curated dataset comprising 20,842 proteins constructed for model development and validation.

Scientific Applications:

• Industrial enzyme discovery: Identification and selection of thermostable enzymes for industrial processes that require high-temperature stability.
• Thermostable biopharmaceutical development: Screening and engineering of proteins for development of thermostable pharmaceuticals and related biotechnological applications.

Methodology:

Data preparation involved curation of a large-scale dataset of 20,842 proteins; feature extraction used CNNs and BiLSTMs to obtain sequence-derived representations; self-attention mechanisms assigned importance to extracted features which were then fused with biological feature channels; model training and validation compared DeepTP to existing methods, reporting an AUC of 0.944 on a balanced validation set, an AUC of 0.801 on another validation set, and an average precision (AP) of 0.536 on an unbalanced test set.