Trainable High Resolution Melt Curve Machine Learning Classifier for Large-Scale Reliable Genotyping of Sequence Variants

Abstract

High resolution melt (HRM) is gaining considerable popularity as a simple and robust method for genotyping sequence variants. However, accurate genotyping of an unknown sample for which a large number of possible variants may exist will require an automated HRM curve identification method capable of comparing unknowns against a large cohort of known sequence variants. Herein, we describe a new method for automated HRM curve classification based on machine learning methods and learned tolerance for reaction condition deviations. We tested this method in silico through multiple cross-validations using curves generated from 9 different simulated experimental conditions to classify 92 known serotypes of Streptococcus pneumoniae and demonstrated over 99% accuracy with 8 training curves per serotype. In vitro verification of the algorithm was tested using sequence variants of a cancer-related gene and demonstrated 100% accuracy with 3 training curves per sequence variant. The machine learning algorithm enabled reliable, scalable, and automated HRM genotyping analysis with broad potential clinical and epidemiological applications.

Arbovirus and insect‐specific virus discovery in Kenya by novel six genera multiplex high‐resolution melting analysis

“…However, reliable scoring of HRM profiles does depend on comparisons to positive controls that might not be readily available for the full panel of potential targets. By developing online database approaches by which HRM data from different real-time PCR platforms and laboratories can be shared, compared and classified using machine learning algorithms (Athamanolap et al 2014), the need for positive controls within laboratories may be reduced. Use of this assay can accelerate identification of novel viruses as they emerge in new geographies, such as Zika virus, a virus previously thought of as relatively benign to humans but has recently spread to the American continents where it is responsible for severe neurological birth defects (Fauci & Morens 2016).…”

“…; Athamanolap et al . ) and malarial parasites (Kipanga et al . ), as well as with low‐level multiplex PCR products to detect and differentiate dengue viruses (Waggoner et al .…”

“…Lambert & Lanciotti (2009) employed triplex PCR for sequencing-based identification of diverse arboviruses in the family Bunyaviridae (Orthobunyavirus, Phlebovirus, Nairovirus). Multiplex PCR-HRM analysis has been performed using universal primers to differentiate bacterial pathogens (Yang et al 2009;Xue et al 2012;Athamanolap et al 2014) and malarial parasites (Kipanga et al 2014), as well as with low-level multiplex PCR products to detect and differentiate dengue viruses (Waggoner et al 2013). With the exception of the dengue typing assay of Waggoner and colleagues (Waggoner et al 2013), reliable HRM methods have been limited to applications in which template quality and quantity could be controlled to generate consistent and comparable profiles, using either cultured viruses (Yang et al 2009) or synthesized templates (Naze et al 2009).…”

Three-gene PCR and high-resolution melting analysis for differentiating vertebrate species mitochondrial DNA for biodiversity research and complementing forensic surveillance

“…Therefore, for comparable and reproducible HRM results, all samples should be processed under the same conditions. However, with more affordable (and probably advanced) HRM-capable thermocyclers entering the market, such as the MIC-4 (Bio Molecular Systems, Australia) and Chai's Open qPCR (Chai, CA, USA) thermocyclers, cross-platform comparisons may be aided by novel algorithms to harmonize HRM analysis across platforms 41 .…”

Machine Learning-Assisted Digital PCR and Melt Enables Broad Bacteria Identification and Pheno-Molecular Antimicrobial Susceptibility Test

“…Following relative Tm-based classification, to enable the identification of digital melt curves of interest based on their shapes, we normalize the area under the curve of all digital melt curves and then align them to a single point. Finally, using an adapted in-house developed digital melt curve classification tool based on one-versus-one support vector machine (ovoSVM) algorithm [26,48,53] , the shape of each digital melt curve of interest is compared to the digital melt curves in the Tm-classified group to identify the bacterial species represented by the digital melt curve of interest.…”