Convolutional neural network analysis of recurrence plots for high resolution melting classification

Area of Science:

• Bioinformatics
• Machine Learning
• Genomics

Background:

• High-resolution melting (HRM) analysis is crucial for species identification across various fields.
• HRM data, generated via real-time PCR, can vary with experimental conditions, complicating analysis.
• Classifying similar species using HRM data is challenging due to data complexity and volume.

Purpose of the Study:

• To enhance the accuracy of species classification using High-resolution melting (HRM) data.
• To develop novel methods for analyzing and classifying HRM data, particularly for species with similar characteristics.
• To investigate the efficacy of image-based representations and deep learning models for HRM data analysis.

Main Methods:

• HRM data was transformed into visual representations termed HRM images using recurrence plots (black-white and grayscale).
• Convolutional neural network (CNN) models were developed and trained to classify these HRM images.
• Performance was compared against traditional melting curve analysis and Support Vector Machine (SVM) models.

Main Results:

• The CNN models utilizing black-white recurrence plot (BW-RP) HRM images achieved the highest classification accuracy of 95.2%.
• BW-RP based CNN models outperformed CNNs using grayscale recurrence plots (86.13%) and melting curve data (90.13%).
• The BW-RP approach also yielded superior F1 scores, specificity, recall, and precision for most species compared to SVM and other CNN models.

Conclusions:

• Image-based representation of HRM data, specifically BW-RP, significantly improves species classification accuracy.
• CNN models trained on BW-RP HRM images demonstrate superior performance over traditional methods and other data representations.
• This approach offers a robust solution for accurate and efficient species identification from HRM data.