Date of Publication
2021
Document Type
Master's Thesis
Degree Name
Master of Science in Statistics
Subject Categories
Mathematics
College
College of Science
Department/Unit
Mathematics and Statistics Department
Thesis Advisor
Frumencio F. Co
Defense Panel Chair
Rechel G. Arcilla
Defense Panel Member
Romeric F. Pobre
Erold Ryan A. Bonzo
Abstract/Summary
Digital PCR (dPCR) is an emerging technology to detect and quantify target DNA sequences for applications such as medical diagnosis, forensic research, and food safety analysis. As the use of dPCR gains more popularity in recent years, each step in its workflow must be improved to surpass its performance over the gold standard real-time qPCR. Its novel approach in partitioning target samples into equal-sized droplets makes it appealing to be more theoretically accurate than qPCR. Droplets containing at least one target DNA emits a high fluorescence intensity and is classified as positive; otherwise, low intensity is emitted and is classified as negative. Classification becomes complicated when several intermediate droplets called "rain" are present, causing severe misclassification. Since nonoptimal data is frequent in dPCR studies, droplet classifiers should be robust to the presence of rain, baseline shifts, multiple fluorescence populations, and poor separation of populations. This thesis reviews the current droplet classification methods of single-channel dPCR quantification, which are Cloudy, ddpcRquant, and Umbrella. The Expectation-Maximization (EM) Clustering is proposed to address plausible research gaps and improve current classification performance in the dataset with samples of varying quality from Lievens et al. (2017) and a simulated dataset. The results show that the proposed method using T- and skewed T-mixture models have mostly outperformed the precision in terms of CV amongst three current methods, and is on par or better in terms of accuracy and linearity of target concentration estimates. Finally, the proposed method is freely available for public use by installing the “popPCR” R package from CRAN (The Comprehensive R Archive Network).
Abstract Format
html
Language
English
Format
Electronic
Physical Description
109 leaves
Keywords
Polymerase chain reaction; Technology; Drops
Recommended Citation
Guiao, J. (2021). Model-based clustering of digital PCR droplets using expectation-maximization. Retrieved from https://animorepository.dlsu.edu.ph/etdm_math/1
Upload Full Text
wf_yes
Embargo Period
5-5-2021
