DeepPPThermo: A Deep Learning Framework for Predicting Protein Thermostability Combining Protein-Level and Amino...

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• Discovering thermophilic proteins and enhancing protein thermostability through traditional experimental methods is inefficient and costly.
• Machine learning (ML) has emerged as a powerful tool for predicting protein thermostability, but effectively utilizing multiview sequence information remains a challenge.

Purpose of the Study:

• To develop a deep learning-based classifier, DeepPPThermo, for accurate prediction of thermophilic and mesophilic proteins.
• To fuse classical sequence features with deep learning representation features for improved thermostability prediction.

Main Methods:

• Proposed DeepPPThermo, a deep learning classifier integrating classical sequence features and deep learning representations.
• Employed deep neural network (DNN) and bi-long short-term memory (Bi-LSTM) to extract hidden features.
• Utilized local and global attention mechanisms to assign differential importance to multiview features, feeding fused features into a fully connected network classifier.

Main Results:

• DeepPPThermo demonstrated superior performance compared to advanced ML and deep learning algorithms in classifying thermophilic and mesophilic proteins.
• Ablation studies confirmed the importance of individual features and the overall robustness of the DeepPPThermo model.

Conclusions:

• DeepPPThermo offers a robust and effective deep learning approach for predicting protein thermostability.
• The model can aid in exploring protein diversity, identifying novel thermophilic proteins, and guiding directed mutations for protein engineering.