2.6. High-Resolution Melting Data Analysis

Values obtained from the DNA fusion curves were analyzed with the High-Resolution Melt software for LightCycler 96 System Software and, to avoid manual analysis of the HRM melting curves, we developed an algorithm in Python based on the algorithm proposed by Li et al. (2016) [30]. We extended the algorithm to allow for domain identification and analysis and analysis against multiple control curves considering the possible deviations of normal curves. Instead of using predefined angles to detect the start and end of the melting region, we used the second derivative of the melting curve to detect minimum points and peaks. The code is available at: https://gitfront.io/r/user-5547184/d3ec1062bc7d0d2e3039b74770330c06f510d65e/hrm-analysis/, accessed on 20 August 2021.

Melting large amplicons, such as exons with a size greater than 300 bp, can result in multiple melting domains, which complicate the HRM analysis. Usually, the solution for this is to create smaller amplicons, but this can increase the cost of the analysis as multiple samples are needed. Instead, we analyzed each domain separately. A domain is identified as a peak in the derivative curve. A peak in a curve is detected when its derivative changes from negative to positive. To identify domains in the melting curve, we applied a Savitsky–Golay filter using the savgol_filter function of the scipy signal library for Python to identify the second derivative of the curve. We then identified the points where it changes its sign from positive to negative. These are considered the peaks in the first derivative of the melting curve. Peaks with values below a threshold were not considered a domain for the analysis, as they were found to be noise. The threshold value was set to 0.10 by analyzing the number of domains in most curves. Each of the resulting peaks represented a domain in the melting curve. For each domain, we then identified its melting region.

To identify the melting region of each domain, we found the minimum points in the second derivative. These are identified as the points where the sign changes from positive to negative. The start of the melting region was set 2 degrees before the start point of the domain and the end of the melting region as 2 degrees after the end of the domain, as suggested by Li et al. (2016) [30]. For each domain, curve normalization and background subtraction were performed following the procedure suggested by Li et al (2016) [30]. The melting temperature corresponded with the peak of the domain found previously.

The preferred data display for high-resolution amplicon melting analysis is a difference plot which shows the difference between the analyzed curve and the mean curve of the set of controls. One difference plot for each domain was created. The median was chosen instead of the mean curve because the median curve is less affected by outliers. We also plotted the standard deviation of the median curve (shown as a gray area) to represent the variability of the set of controls.