Incorporating support vector machine with sequential minimal optimization to identify anticancer peptides.

Area of Science:

• Biochemistry
• Computational Biology
• Bioinformatics

Background:

• Cancer remains a leading global cause of death.
• Anticancer peptides (ACPs) offer a promising targeted therapy, but their identification is challenging.
• Existing computational tools for ACP prediction have limitations, particularly in differentiating ACPs from antimicrobial peptides (AMPs).

Purpose of the Study:

• To develop advanced computational models for predicting anticancer peptides (ACPs).
• To create a model capable of distinguishing ACPs from antimicrobial peptides (AMPs).
• To enhance the accuracy and utility of computational tools for ACP identification.

Main Methods:

• Utilized features including amino acid composition, N5C5, k-space, and Position-Specific Scoring Matrix (PSSM).
• Employed machine learning algorithms such as Support Vector Machine (SVM) and Sequential Minimal Optimization (SMO).
• Developed two models: one for distinguishing ACPs from AMPs, and another for predicting ACPs from all peptides.

Main Results:

• Achieved high performance with an accuracy of 85.5% for the ACP vs. AMP model (model 1).
• Demonstrated superior performance with 95.2% accuracy for the ACP prediction model (model 2).
• The SMO-optimized model using PSSM outperformed all existing ACP prediction tools.

Conclusions:

• The Position-Specific Scoring Matrix (PSSM) feature significantly improves prediction performance.
• Sequential Minimal Optimization (SMO) enhanced Support Vector Machine (SVM) models, leading to better results.
• This research provides valuable computational tools for distinguishing ACPs from AMPs and all peptides, with the PSSM-based SMO model showing state-of-the-art performance.