(I-348) Improving Accuracy of Digital Melt via Oligonucleotide-Enabled Curve Alignment

Introduction:Polymicrobial bacterial infections of two or more causative bacterial species, which account for up to 14% of bloodstream infections and result in worsened patient outcomes¹, are difficult to diagnose by conventional diagnostic tools. Digital melt is an emerging method for diagnosing polymicrobial bacterial infections by first isolating individual bacterial genomes in microfluidic reaction wells and then generating bacterial species-specific melt curves for identification. However, existing digital melt platforms are susceptible to inaccuracies due to nonuniform digital melt curves, which are caused by unwanted temperature gradients from imperfect heating. Thus, there is a pressing need for a curve calibration method to improve the accuracy of digital melt. Herein, we introduce a new method that uses a DNA sequence with a known melting temperature to facilitate curve alignment, minimize heating-induced curve nonuniformities, and improve identification accuracy.


Materials and Methods: Digital melt was performed on a microfluidic nanoarray² using a PCR mixture containing PCR primers that target the 16S V1 – V6 region either with or without the DNA calibrator. Calibration of the bacterial melt curves was performed by translating each curve by the average shift needed to align the calibrator peaks in surrounding negative wells multiplied by the ratio of the bacterial T_M and the calibrator T_M standard deviations (Figure 1a). To evaluate curve alignment, digital melt curves from A. baumannii were generated and analyzed (Figure 1b). To evaluate identification accuracy, digital melt curves for K. pneumoniae and M. morganii, whose melting temperatures differed by < 1 ºC, were generated to train a machine learning identification algorithm, which was subsequently used to identify the two species in a 1:1 mixture.


Results, Conclusions, and Discussions: Results and Discussion: We validated that calibration improves alignment of melt curves for three bacterial species. Results for A. baumannii are pictured in Figure 1b, which shows that calibration visually improves peak alignment. The heatmap of the bacterial T_Ms shows the pre-calibration temperature gradient across the module is reduced post-calibration, verifying that calibration reduces the effects of cross-platform heating variations. Overall, calibration reduces the T_M standard deviation by 64.3% for A. baumannii, 33.5% for K. pneumoniae, and 30.6% for M. morganii. Additionally, calibration improved our machine learning algorithm’s ability to distinguish between K. pneumoniae and M. morganii in a 1:1 mixture. Before calibration, the algorithm identified the 1:1 mixture as a 23.4:1 mixture. After calibration, identification improved, with the algorithm classifying the sample as a 2.1:1 mixture.

Conclusions: We developed and validated a method to align digital melt curves using an internal DNA calibrator. Our results show improved melt curve alignment for three infection-causative bacteria and increased bacterial identification accuracy for a polymicrobial mixture. These results suggest that this calibration method can improve identification of polymicrobial bacterial infections using nanoarray digital melt.


References : [1] Doualeh et al., Int. J. Mol. Sci. 2022; 23(9): 4484. [2] Athamanolap et al., Anal. Chem. 2019; 91(20): 12784-12792.