Novel signal processing and classification methods for forensic species identification

  • Sorelle Jean Bowman

Student thesis: Doctoral Thesis


Non-human species identification from animal and bacterial origin, is an important
aspect of forensic investigations and relates to criminal matters, food safety and
security, wildlife forensics and biosecurity. Animal-derived biological sources take
many forms, for example, illegal powdered animal parts used in medicines,
adulterated meat products, blood of unknown unknown origin taken from a vehicle
involved in a hit-and-run case, transferred pet hair on a suspects clothing. Many types
of non-human material can represent valuable evidence in casework. Furthermore,
bacterial species of focus in this study are biological agents that pose a concern to
national security if deliberately disseminated in an act of bioterrorism, specifically
Bacillus anthracis and Yersinia pestis which cause anthrax and plague, respectively.
An ideal method for the detection and differentiation of both animal and bacterial
species should demonstrate characteristics such as high sensitivity, specificity,
rapidity, cost-effectiveness, simplicity and should ideally be able to be employed in a
field setting.
The aim of this thesis was to detect and identify species of forensic interest,
differentiating them from each other and from potential hoax agents. This broad aim
could be further divided as follows; (1) Examine a range of DNA extraction
procedures with a view to identifying a fast and efficient method that removes PCR
inhibitors, ideally suited for use in the field. (2) Examine the use of high resolution
melting (HRM) analysis for its ability to differentiate between species of forensic
interest, including animals and bacteria (biothreat agents). (3) Examine the use of
targeted massively parallel sequencing (MPS) to identify biothreat agents and
compare this with quantitative real time PCR (qPCR). (4) Employ novel signal
processing and species classification methods for application to microfluidic capillary
electrophoresis (MCE) of proteins, HRM and MPS. The aims span two genetic targets
(proteins, DNA), four species detection methods (MCE, qPCR, HRM and MPS) and
four classification methods (peak detection algorithms, Boolean logic gates,
classification trees and MPS sequence alignment). They were combined in a number
of different permutations to provide a suite of forensic species identification solutions Four commercial DNA extraction methods were applied to both Gram-negative
(Bacillus species) and Gram-positive (E. coli) bacterial cultures and were evaluated
for their application in matters of biosecurity. These were ChargeSwitch gDNA mini
bacteria kit (Invitrogen), QIAamp DNA extraction kit (Qiagen) with and without
bead-beating, and Isolate II Genomic DNA kit (Bioline). The Isolate II Genomic
DNA kit was found to remove inhibitors cost effectively for the extraction of bacterial
DNA from both culture and environmental samples.
The universal 16S rRNA gene was targeted with HRM and used to generate
derivative melt profiles for human and ten animal species typically encountered in
forensic case work, as either consumed meats (Gallus gallus (chicken), Bos taurus
(cow), Sus scrofa (pig) and Ovis aries (sheep)), domestic pets (Felis catus (cat) and
Canis lupus familiaris (dog)) or Australian road kill (Vulpes vulpes (fox), Macropus
(kangaroo), Vombatus ursinus (wombat) and Oryctolagus cuniculus (rabbit)). HRM
derivative melt profiles were processed and analysed by random forest classification
(such as classification trees, bagging and boosting) and peak detection algorithms
with Boolean logic paths. Random forest classification, particularly bagging, was the
most suitable for the purposes of animal species identification with a prediction
accuracy of 90.8 % for the randomly partitioned test dataset and 70 % for the
validation dataset across all species.
Microfluidic capillary electrophoresis and HRM were evaluated for their use as a
screening tool for bacterial species identification with a focus on Bacillus and
Yersinia species, E. coli and powder-based hoax agents (e.g. Dipel and plain wheat
flour). The signals generated from both platforms were characterised by peak
detection algorithms and differentiated using Boolean logic paths. When applied to
protein profiling by MCE, peak detection and classification by Boolean logic yielded
predictive accuracy of 75 % with the test dataset, across all samples. Additionally,
when the same algorithmic approach was applied to HRM derivative melt profiles, the
seven Bacillus species could be differentiated into B. cereus group members and non-
B. cereus group members MPS was used to develop a targeted sequencing approach to identify Bacillus species,
in particular B. anthracis, in samples collected at the Canberra Airport. Two virulence
plasmid markers (cya and capB) and a single chromosomal marker (16S rRNA gene)
were targeted to establish background B. anthracis frequencies over a 12 month
sampling period (from August 2011 to July 2012). The findings demonstrated
effective reference alignment to define bacterial entry, dispersal and movement
throughout the airport. Of the 20 samples sequenced, 15 were positive for the
B. cereus group 16S rRNA gene, two samples were cya positive in the month of
February 2012 and seven were capB positive in the sampling months of December
2011 and June 2012. A total of four samples collected in the sampling month of
February, 2012, were positive for all three markers, indicative of the potential
presence of B. anthracis.
In summary, this study has resulted in novel screening approaches, across multiple
platforms, for the effective detection and analyses of both animal and bacterial species
for forensic purposes. Moreover, the analysis has provided a foundation for future
work involving targeted sequencing of the bacterial metagenomic background of a
public transport hub and the movement of B. anthracis outside the anthrax belt.
Date of Award2018
Original languageEnglish
SupervisorMichelle Gahan (Supervisor) & Dennis Mcnevin (Supervisor)

Cite this