Advanced search
2 files | 228.74 MB

Confronting bias and precision in digital PCR quantification

Bart Jacobs (UGent)
(2017)
Author
Promoter
(UGent) and (UGent)
Organization
Abstract
Nucleic acid (NA) sequences such as DNA are often used in the life sciences, e.g. for identifying and characterising infections, cancer and genetic defects in the field of medicine, or quantifying and detecting genetically modified organisms (GMOs) or harmful allergens in food and feed. These applications typically require the accurate estimation of one or more target sequences. Digital polymerase chain reaction (dPCR) is an increasingly popular technique to detect or estimate the concentration of such target NA sequences in a sample. Manufacturers claim it is more performant than classical methods and published results by early adopters were highly encouraging. Nevertheless, like all technologies, dPCR is subject to known sources of error and variation that may drastically affect the results. Publication guidelines of dPCR data nowadays highlight the importance of recognising essential sources of variation in the measurement process, many of which were initially ignored. Neglecting these sources of error could however lead to misleadingly high expectations and subsequent disappointment for prospective users who count on superior accuracy to be in line with regulation or standards set by other technology. We initiated our research by studying the impact of adapting specific machine settings on the reported concentration estimates and their associated variation by means of a simulation study. This allowed to estimate the isolated effect of different sources of variation in a controlled setting, and recognise the most critical steps in a typical dPCR workflow. This input can help informed users in the design and correct interpretation of their experiments. Once we identified factors with important impact on the effect size, we developed a probabilistic model for the downstream data-analysis of dPCR output that incorporates various sources of technical variation in an objective, data-driven way. This model was extensively tested on simulated data and ultimately used to analyse a broad spectrum of typical single-channel and multi-channel datasets. Our model incorporates a natural measure of variation which is missing in standard manufacturer software output, and offers users a wide range of output which can aid in interpretation of the main results, as well as provide some level of quality control.
Keywords
statistics, biostatistics, generalized additive modeling, gam, digital PCR, dPCR, bias, precision, accuracy, variance, replication, detection, quantification

Downloads

  • (...).pdf
    • full text
    • |
    • UGent only
    • |
    • PDF
    • |
    • 101.41 MB
  • (...).pdf
    • full text
    • |
    • UGent only
    • |
    • PDF
    • |
    • 127.33 MB

Citation

Please use this url to cite or link to this publication:

Chicago
Jacobs, Bart. 2017. “Confronting Bias and Precision in Digital PCR Quantification”. Ghent, Belgium: Ghent University. Faculty of Sciences.
APA
Jacobs, Bart. (2017). Confronting bias and precision in digital PCR quantification. Ghent University. Faculty of Sciences, Ghent, Belgium.
Vancouver
1.
Jacobs B. Confronting bias and precision in digital PCR quantification. [Ghent, Belgium]: Ghent University. Faculty of Sciences; 2017.
MLA
Jacobs, Bart. “Confronting Bias and Precision in Digital PCR Quantification.” 2017 : n. pag. Print.
@phdthesis{8535408,
  abstract     = {Nucleic acid (NA) sequences such as DNA are often used in the life sciences, e.g. for identifying and characterising infections, cancer and genetic defects in the field of medicine, or quantifying and detecting genetically modified organisms (GMOs) or harmful allergens in food and feed.  These applications typically require the accurate estimation of one or more target sequences. Digital polymerase chain reaction (dPCR) is an increasingly popular technique to detect or estimate the concentration of such target NA sequences in a sample. Manufacturers claim it is more performant than classical methods and published results by early adopters were highly encouraging.
Nevertheless, like all technologies, dPCR is subject to known sources of error and variation that may drastically affect the results. Publication guidelines of dPCR data nowadays highlight the importance of recognising essential sources of variation in the measurement process, many of which were initially ignored. Neglecting these sources of error could however lead to misleadingly high expectations and subsequent disappointment for prospective users who count on superior accuracy to be in line with regulation or standards set by other technology.
We initiated our research by studying the impact of adapting specific machine settings on the reported concentration estimates and their associated variation by means of a simulation study. This allowed to estimate the isolated effect of different sources of variation in a controlled setting, and recognise the most critical steps in a typical dPCR workflow. This input can help informed users in the design and correct interpretation of their experiments.
Once we identified factors with important impact on the effect size, we developed a probabilistic model for the downstream data-analysis of dPCR output that incorporates various sources of technical variation in an objective, data-driven way. This model was extensively tested on simulated data and ultimately used to analyse a broad spectrum of typical single-channel and multi-channel datasets. Our model incorporates a natural measure of variation which is missing in standard manufacturer software output, and offers users a wide range of output which can aid in interpretation of the main results, as well as provide some level of quality control.},
  author       = {Jacobs, Bart},
  language     = {eng},
  pages        = {XX, 192},
  publisher    = {Ghent University. Faculty of Sciences},
  school       = {Ghent University},
  title        = {Confronting bias and precision in digital PCR quantification},
  year         = {2017},
}