Improving Prediction of Self-interacting Proteins Using Stacked Sparse Auto-Encoder with PSSM profiles.

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in protein science

Background:

• Self-interacting proteins (SIPs) are vital for cellular functions including signal transduction and gene regulation.
• Experimental methods for SIP identification are costly and time-consuming, necessitating efficient computational approaches.

Purpose of the Study:

• To develop and validate a novel computational method for accurate identification of self-interacting proteins (SIPs).
• To leverage machine learning techniques for efficient feature extraction and classification of SIPs.

Main Methods:

• Utilized Zernike Moments (ZMs) for feature extraction from Position Specific Scoring Matrix (PSSM) data.
• Employed Stacked Sparse Auto-Encoder (SSAE) for deep feature dimension reduction and noise removal.
• Applied Probabilistic Classification Vector Machines (PCVM) for the final classification of SIPs.

Main Results:

• The proposed method achieved high prediction accuracies of 92.55% on *S. cerevisiae* and 97.47% on Human SIP datasets.
• Comparative analysis demonstrated that the PCVM classifier outperformed Support Vector Machine (SVM) and other existing methods.
• The integrated approach shows significant promise for accurate and efficient SIP identification.

Conclusions:

• The novel computational strategy integrating ZMs, SSAE, and PCVM offers a powerful tool for identifying self-interacting proteins.
• This machine learning-based approach provides a cost-effective and time-efficient alternative to experimental methods for SIP detection.
• The study highlights the potential of advanced computational techniques in advancing our understanding of protein interactions and cellular mechanisms.