Back to search

Georgia Institute of Technology

Probabilistic Methods For Rapid Characterization Of Complex Biological Samples

Abstract

dc:description.abstract

A descriptive set of statistics metrics and machine learning models are developed toboth discriminate and classify bacteria using high throughput and information rich flow cytometry, as well as automate and improve a novel colorimetric antibiotic susceptibility testing (AST) assay, ChroMIC. Bayesian classifiers are utilized to provide a fast, flexible platform to perform rapid bacterial identification of not only unknown samples with one causative agent, but this analysis makes good on the ability to classify binary mixtures as well. Model training and validation are achieved by first constructing a library of known bacterial pathogens, then using the models to predict bacterial identity of unknown pathogens. These Bayesian methods are coupled to our previously described Probability Binning Signature Quadratic Form (PB-sQF) statistics, which enable further rigor. A fast Gram-stain support vector machine (SVM) classifier is trained with a commercially available dye kit, and was developed to pair easily with AST workflows currently in development. Once trained, the SVM classifier allows for the determination of the Gram status of unknown bacterial pathogens within minutes of bacterial isolation. To further refine and speed up the AST step of the clinical workflow, the previously developed colorimetric AST ChroMIC was automated and refined to be a real-time assay. This simple colorimetric assay uses machine learning and computer vision to auto label bacterial growth positive and negative wells with a modified, self-correcting SVM model. ChroMIC has high categorical and essential agreements (> 90%) with the clinical gold standard technique of broth micro dilution, meaning that ChroMIC is a strong candidate for future deployment in clinical settings. These described assays and methods constitute a complete set of clinically relevant technologies, meaning that both the identity and the antibiotic susceptibility profile of the unknown pathogen are determined rapidly post blood-culture.

Degree

thesis:*
Level thesis:degree_level
Doctoral
Department dc:contributor.department
Chemistry and Biochemistry
Grantor dc:publisher
Georgia Institute of Technology
Year dc:date.issued
2022

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Richardson, Joseph Clayton
Advisor dc:contributor.advisor
  • Dickson, Robert M.
Committee members dc:contributor.committeemember
  • Fahrni, Christoph
  • Cicerone, Marcus
  • Hud, Nicholas

Subjects

dc:subject × 6

Rights

Language dc:language.iso
en_US

Identifiers

dc:identifier.*
Handle dc:identifier.uri
https://hdl.handle.net/1853/73060

Chain of custody

source
Harvested from
Georgia Tech
Base URL
repository.gatech.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Richardson, Joseph Clayton. Probabilistic Methods For Rapid Characterization Of Complex Biological Samples. Doctoral thesis, Georgia Institute of Technology, 2022. https://hdl.handle.net/1853/73060