DeepSaltPro: Enhancing halophilic protein prediction accuracy and efficiency via multi-protein language model...

Area of Science:

• Biochemistry
• Computational Biology
• Artificial Intelligence

Background:

• Halophilic proteins are vital for high-salt environments, with applications in industry and biotechnology.
• Experimental identification of these proteins is challenging and inefficient for large-scale studies.
• Computational methods are needed for accurate and efficient halophilic protein prediction.

Purpose of the Study:

• To develop DeepSaltPro, a deep learning framework for predicting halophilic proteins.
• To leverage pre-trained protein language models and advanced neural network architectures.
• To provide an efficient computational tool for identifying halophilic proteins.

Main Methods:

• Feature extraction using Ankh and ESM-2 protein language models.
• Integration of Convolutional Neural Networks (CNNs) and Bidirectional Gated Recurrent Units (BiGRU).
• Utilizing Kolmogorov-Arnold Networks (KAN) for complex nonlinear interactions and interpretability.

Main Results:

• DeepSaltPro achieved 97% accuracy on an independent test set.
• Demonstrated a 12% improvement over the existing state-of-the-art method, HPClas.
• Validated performance through five-fold cross-validation and independent testing.

Conclusions:

• DeepSaltPro is an effective computational tool for identifying halophilic proteins.
• The framework offers valuable insights into protein function and mechanisms.
• Facilitates future industrial and biotechnological applications of halophilic proteins.