Discriminant analysis and random forests are used to create models for classification. The number of variables to be tested for inclusion in a model can be large. The goal of this work was to create an efficient and effective selection program. The first method used was based on the work of others. The resulting models were underperforming, so another approach was adopted. Models were built by adding the variable that maximized new-model accuracy. The two programs were used to generate discriminant-analysis and random forest models for three data sets. An existing software package was also used. The second program outperformed the alternatives. For the small number of runs produced in this study, it outperformed the method that inspired this work. The data sets were studied to identify determinants of performance. No definite conclusions were reached, but the results suggest topics for future study.

This thesis addresses the genetic characterisation of livestock breeds, a key aspect of the long-term future breed preservation and, thus, of primary interest for animal breeders and management in the industry. First, the genetic diversity and structure of breeds were investigated. The application of individual-based population genetic approaches at characterising genetic structure was assessed using the British pig breeds. All approaches, except for Principle Component Analysis (PCA), found that the breeds were distinct genetic populations. Bayesian genotypic clustering tools agreed that breeds had little individual genetic admixture. However, inconsistent results were observed between the Bayesian methods. Primarily, BAPS detected finer genetic differentiation than other approaches, producing biologically credible genetic populations. BAPS also detected substructure in the British Meishan, consistent with prior known population information. In contrast, STRUCTURE detected substructure in the British Saddleback breed that could not wholly be explained. Further analysis of the British Saddleback revealed that the genetic subdivision did not reflect its historical origin (union of Essex pig and Wessex Saddleback) but was associated with herds. The Rainbarrow appeared to be moderately differentiated from the other herds, and relatively lower allelic diversity and higher individual inbreeding, a possible result of certain breeding strategies. The genetic structure and diversity of the British traditional chicken breeds was also characterised. The breeds were found to be highly distinctive populations with moderately high levels of within-breed genetic diversity. However, majority of the breeds had an observed heterozygote deficit. Although individuals clustered to their origin for some of the breeds, genetic subdivision of individuals was observed in some breeds. For two breeds the inferred genetic subpopulations were associated with morphological varieties, but in others they were associated with flock supplier. As with the British Saddleback breed, gene flow between flocks within the chicken breeds should be enhanced to maintain current levels of genetic diversity. Second, the thesis focused on breed identification through the assignment of individuals to breed origin. Dense genome-wide assays provide an opportunity to develop tailor-made panels for food authentication, especially for verifying traditional breed-labelled products. In European cattle breeds, the prior selection of informative markers produced higher correct individual identification than panels of randomly selected markers. Selecting breed informative markers was more powerful using delta (allele frequency difference) and Wright's FST (allele frequency variation), than PCA. However, no further gain in power of assignment was achieved by sampling in excess of 200 markers. The power of assignment and number of markers required was dependent on the levels of breed genetic distinctiveness. Use of dense genome-wide assays and marker selection was further assessed in the British pig breeds. With delta, it was found that 96 informative SNP markers were sufficient for breed differentiation, with the exception of Landrace and Welsh pair. Assignment of individuals to breed origin was high and few individuals were falsely assigned, especially for the traditional breeds. The probability that a sample of a presumed origin actually originated from that breed was high in the traditional breeds. Validation of the 96-SNP panel using independent test samples of known origin and market samples revealed a high level of breed label conformity.

>Magister Scientiae - MSc / Aims DNA metabarcoding has recently emerged as a valuable supplementary tool to ensure food authenticity within the global food market. However, it is widely known that highly processed food samples are one of DNA metabarcoding’s greatest shortfalls due to high DNA degradation, presence of PCR inhibitors and the incomplete removal of several undesirable compounds (such as polysaccharides) that makes the amplification of desired DNA challenging. This project has two main aims, the first of which was to determine and develop a cost and time effective DNA metabarcoding system that could successfully describe to species level the ingredient composition of highly processed vegetarian food products. The DNA metabarcoding system was thoroughly evaluated and tested by combining well-researched primers with varying concentrations into a multiplex reaction. The combination of plant and animal primers selected that yielded the best results were used to determine the species composition in the samples. The second aim is to determine the possible presence of meat contaminants within the highly processed vegetarian food samples. Numerous studies have shown that food adulteration is a wide-spread phenomenon throughout the world due to the economic gains it can provide. Animal primers were introduced into the multiplex reaction to aid in the identification of any meat products that could have been inserted into the vegetarian products to lower the overall cost to company. Methodology Thirty-two highly processed vegetarian food samples were collected in the Cape Town area from local and franchised supermarkets. DNA was extracted using the Chloroform/Isoamyl alcohol method best suited for plant-based samples followed by amplification of the following mini-barcoding regions: the mitochondrial 16S ribosomal rRNA, cytochrome B, tRNALeu – trnL – UAA intron and the ribosomal internal transcribed spacer region – ITS2 for plant and fungi identification. The PCR products were purified using the Qiaquick kit and library preparation and building was conducted using the TruSeq DNA PCR-free Library kit. Final purification was completed using AMPure XP kit and the pooled libraries were sequenced on an Illumina Miseq using 300bp paired-end run. Statistical and bioinformatic analysis on the NGS raw sequence reads was performed in R version 3.6.3. Results The results of the data analysis showed that the cytochrome B primer couldn’t detect any animal DNA in the vegetarian samples, however animal-derived sequences were detected in the positives present, validating the efficacy of the multiplex reaction. Mitochondrial 16S ribosomal rRNA was only able to detect plant-based DNA due to the structural homology between chloroplast and mitochondrial DNA. The fungal ribosomal internal transcribed spacer region – ITS2 detected sequences deriving from “Viridiplantae”. This result could have been due to the fungal and plant ribosomal internal transcribed spacer region – ITS2 sharing a reverse primer during amplification. The trnL region was able to detect the presence of undeclared coriander, mustard and wheat in 8 (29%), 6 (21%) and 5 (18%) samples respectively. Additionally, trnL was able to detect the presence of tobacco in 11 (35%) samples. This could have been due to cross-contamination between samples being co-extracted and amplified at the same time for separate studies. The PITS2 region was able to detect the presence of undeclared barley, mustard and wheat in 8 (25%), 4 (14%) and 4 (14%) samples respectively. Our results show the possibility of DNA metabarcoding for the authentication of a wide range of species present in highly processed vegetarian samples using a single assay. However, further optimization of the technique for the identification of both plant and animal species within vegetarian samples needs to be performed before the wide-spread implementation of this technology would be both feasible and viable. Eliminating primer biases, decreasing the risk of homology between different primers in the same assay as well as preventing the amplification of sequencing of undesirable DNA need to be further explored and ultimately mitigated before DNA metabarcoding can be widely seen as an effective and cost-effective method for authentication and food control.

Submitted by Angela Maria de Oliveira (amolivei@uepg.br) on 2019-03-11T20:45:03Z No. of bitstreams: 2 license_rdf: 811 bytes, checksum: e39d27027a6cc9cb039ad269a5db8e34 (MD5) Pablo Monteiro.pdf: 2511532 bytes, checksum: 27e5052e6b53a2fa362e4fdb08bb47c0 (MD5) / Made available in DSpace on 2019-03-11T20:45:03Z (GMT). No. of bitstreams: 2 license_rdf: 811 bytes, checksum: e39d27027a6cc9cb039ad269a5db8e34 (MD5) Pablo Monteiro.pdf: 2511532 bytes, checksum: 27e5052e6b53a2fa362e4fdb08bb47c0 (MD5) Previous issue date: 2018-12-12 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / O café é uma das commodities mais importantes no mundo, sendo o Brasil o maior produtor e exportador do grão (Coffea arabica e Coffea canephora). Os cafés brasileiros são reconhecidos por sua alta qualidade sensorial e pelas propriedades estimulantes. A composição química do café é influenciada por vários fatores, como a altitude em que a planta é cultivada, tipos de secagem do grão, grau de torra a que os grãos são expostos, o sistema de cultivo empregado (orgânico ou convencional), entre outros. O mercado cafeeiro valoriza produtos com sistema de cultivo e local de produção autenticados. Desta forma, o objetivo geral do trabalho foi avaliar o efeito do sistema de cultivo, origem geográfica e origem botânicas de cafés brasileiros na composição fenólica, características físico-químicas e propriedade antioxidante dos grãos. Foram utilizados um total de 45 cafés brasileiros provenientes de Minas Gerais (MG; n = 13), São Paulo (SP; n = 11), Paraná (PR; n = 8), Espírito Santo (ES; n = 3), Bahia (BA; n = 2), e blends: PR/MG/SP (n = 1), MG/SP (n = 6), PR/ES/Roraima (RO) (n = 1). Para avaliar os efeitos dos sistemas de cultivo, foram utilizados n = 19 orgânicos (ORG) e n = 26 convencionais (CONV), sendo que cafés de Coffea arabica n = 41 e blends n = 4 foram estudados em relação à origem botânica. Os resultados da estatística inferencial mostraram que a capacidade de quelar Fe2+ , teor de ácido cafeico, e pH foram diferentes entre as regiões produtoras, sendo que a análise por componentes principais (PCA) não mostrou separação nítida dos cafés de origens geográficas distintas. A análise discriminante por mínimos quadrados parciais (PLSDA) classificou corretamente apenas as amostras do Paraná e blends. O sistema de cultivo (ORG e CONV) influenciou significativamente (p<0,05) a composição fenólica e atividade antioxidante dos cafés, de modo que as amostras ORG apresentaram menores teores de quercetina-3-rutinosídeo, atividade antioxidante medida pelos métodos FRAP e quelar Fe2+, e menores teores de fenólicos totais. A PCA separou os dois grupos efetivamente, ao passo que o modelo de PLS-DA classificou os sistemas de cultivo com eficácia acima de 90%. Em relação à origem botânica, apenas o teor de cafeína mostrou-se diferente entre C. arabica e blends, o que tornou a perfeita classificação da origem botânica das amostras de café possível por PLS-DA. Conclui-se que a utilização de metabolômica referente aos constituintes químicos, atividade antioxidante e propriedades físico-químicas podem ser usadas como marcadores para avaliação da origem botânica, geográfica e do sistema de cultivo de cafés brasileiros. Esses dados são de interesse das indústrias e de órgãos governamentais vistos que autenticação de alimentos é de extrema importância para se ter um produto livre de fraudes, comercialmente competitivo e seguro ao consumo. / Coffee is one of the most important commodities in the world, Brazil being the largest producer and exporter of the grain (Coffea arabica and Coffea canephora). Brazilian coffees are recognized for their high sensory quality and stimulating power. The chemical composition of the coffee is influenced by several factors, such as the altitude at which the plant is grown, types of drying of the grain, degree of roasting in which the grains are exposed, the cultivation system used (organic or conventional), among others. The coffee market values products with certified cultivation system and place of production. In this way, the general objective of the work was to evaluate the effect of the system of cultivation, geographic origin, and botanical origin of Brazilian coffees in the phenolic composition, physical-chemical characteristics, and antioxidant properties. A total of 45 Brazilian coffees from Minas Gerais (MG, n = 13), São Paulo (SP; n = 11), Paraná (PR; n = 8), Espírito Santo (N = 2), and blends: PR/MG/SP (n = 1), MG/SP (n = 6), PR/ES / Roraima (RO) (n = 1). In order to evaluate the effects of the cultivation systems, n = 19 organic (ORG) and n = 26 conventional (CONV) were used. Coffea arabica coffees n = 41 and blends n = 4 were studied in relation to the botanical origin. The results of the inferential statistics showed that the ability to chelate Fe2+ , caffeic acid content, and pH were different among the producing regions, and the principal component analysis (PCA) did not show a clear separation of the coffees from different geographic origins. Partial least squares discriminant analysis (PLS-DA) classified only the Paraná and blends samples. The cultivation system (ORG and CONV) influenced significantly (p<0.05) the antioxidant activity and phenolic composition of the coffee, so that ORG samples showed lower levels of quercetin-3- rutinoside, antioxidant activity measured by FRAP and chelation of Fe2+, and lower total phenolic contents. PCA separated the two groups effectively, while the PLS-DA model ranked cultivation systems effectively above 90%. In relation to the botanical origin, only the caffeine content was different between C. arabica and blends, which made the botanical origin classification of the coffee samples possible (100% efficacy) by PLS-DA. It is concluded that the use of metabolomics in relation to chemical constituents, antioxidant activity and physico-chemical properties can be used as markers for the evaluation of the botanical, geographical origins and the cultivation system of Brazilian coffee. These data are of interest to industries and government agencies that food authentication is of the utmost importance in order to have a fraudfree, commercially competitive and consumer-safe product.