Support vector machine (SVM) based multiclass prediction with basic statistical analysis of plasminogen activators.

Area of Science:

• Biochemistry
• Computational Biology
• Bioinformatics

Background:

• Plasminogen (Pg) is the precursor to plasmin (Pm), a key enzyme in blood proteolysis and fibrinolysis.
• Plasminogen activators (PAs) convert Pg to Pm and are vital for thrombolysis.
• Identifying novel PAs is crucial for understanding their mechanisms and developing new therapies.

Purpose of the Study:

• To investigate computational methods for accurately predicting plasminogen activator peptide sequences.
• To evaluate the performance of Support Vector Machines (SVM) using various sequence features.
• To develop a predictive tool for classifying prokaryotic and eukaryotic plasminogen activators.

Main Methods:

• Employed Support Vector Machines (SVM) with amino acid composition (AC), dipeptide composition (DC), PSSM profiles, and hybrid methods.
• Utilized five-fold cross-validation for performance evaluation.
• Cross-checked prediction accuracy using confusion matrix and ROC analysis.

Main Results:

• Achieved maximum prediction accuracies of 88.37% (AC), 84.32% (DC), 87.61% (PSSM), and 85.63% (Hybrid).
• Demonstrated high correlation among subfamilies of Pg-activators based on sequence compositions.
• Confirmed that plasminogen activators are predictable with high accuracy from primary sequences.

Conclusions:

• Dipeptide, PSSM profile, and Hybrid-based methods outperform single amino acid composition (AC) for Pg-activator prediction.
• A web server for predicting and classifying Pg-activators from primary sequence data has been developed and is publicly available.
• The developed computational approaches are efficient and provide good prediction performance.