A predictive approach for host-pathogen interactions using deep learning and protein sequences.

Area of Science:

• Computational biology
• Bioinformatics
• Infectious disease research

Background:

• Host-pathogen interactions (HPIs) are vital for understanding infections.
• Experimental HPI prediction methods are costly and time-consuming.
• Computational approaches offer efficient alternatives for HPI prediction.

Purpose of the Study:

• To develop a deep learning-based computational method for predicting HPIs using protein sequences.
• To evaluate the accuracy and robustness of the proposed method.

Main Methods:

• Utilized deep learning models for HPI prediction.
• Employed the monoMonoKGap (mMKGap) algorithm (K=2) for feature extraction from protein sequences.
• Generated negative interactions using the Negatome Database.
• Validated the method on three balanced human-pathogen datasets using 10-fold cross-validation.

Main Results:

• Achieved high prediction accuracies: 99.65%, 99.52%, and 99.66% (mean accuracy of 99.61%).
• Demonstrated superior performance compared to Random Forest (RF), Support Vector Machine (SVM), and Convolutional Neural Network (CNN).
• Showcased the effectiveness of the mMKGap feature extraction method over Dipeptide Composition.

Conclusions:

• The proposed deep learning method is highly accurate, robust, and practical for predicting HPIs.
• This framework provides a reliable computational tool for advancing HPI research and combating infectious diseases.