Unsupervised Bayesian Clustering Research Articles

Biases of STRUCTURE software when exploring introduction routes of invasive species.

Population genetic methods are widely used to retrace the introduction routes of invasive species. The unsupervised Bayesian clustering algorithm implemented in STRUCTURE is amongst the most frequently used of these methods, but its ability to provide reliable information about introduction routes has never been assessed. We simulated microsatellite datasets to evaluate the extent to which the results provided by STRUCTURE were misleading for the inference of introduction routes. We focused on an invasion scenario involving one native and two independently introduced populations, because it is the sole scenario that can be rejected when obtaining a particular clustering with a STRUCTURE analysis at K = 2 (two clusters). Results were classified as "misleading" or "non-misleading". We investigated the influence of effective size, bottleneck severity and number of loci on the type and frequency of misleading results. We showed that misleading STRUCTURE results were obtained for 10% of all simulated datasets. Our results highlighted two categories of misleading output. The first occurs when the native population has a low level of diversity. In this case, the two introduced populations may be very similar, despite their independent introduction histories. The second category results from convergence issues in STRUCTURE for K = 2, with strong bottleneck severity and/or large numbers of loci resulting in high levels of differentiation between the three populations. Overall, the risk of being misled by STRUCTURE in the context of introduction routes inferences is moderate, but it is important to remain cautious when low genetic diversity or genuine multimodality between runs are involved.

Genetic variation and population structure of Botswana populations as identified with AmpFLSTR Identifiler short tandem repeat (STR) loci

Population structure was investigated in 990 Botswana individuals according to ethno-linguistics, Bantu and Khoisan, and geography (the nine administrative districts) using the Identifiler autosomal microsatellite markers. Genetic diversity and forensic parameters were calculated for the overall population, and according to ethno-linguistics and geography. The overall combined power of exclusion (CPE) was 0.9999965412 and the combined match probability 6,28 × 10−19. CPE was highest for the Khoisan Tuu ethnolinguistic group and the Northeast District at 0.9999582029 and 0.9999922652 respectively. CMP ranged from 6.28 × 10−19 (Khoisan Tuu) to 1,02 × 10−18 (Northwest district). Using pairwise genetic distances (FST), analysis of molecular variance (AMOVA), factorial correspondence analysis (FCA), and the unsupervised Bayesian clustering method found in STRUCTURE and TESS, ethno-linguistics were found to have a greater influence on population structure than geography. FCA showed clustering between Bantu and Khoisan, and within the Bantu. This Bantu sub-structuring was not seen with STRUCTURE and TESS, which detected clustering only between Bantu and Khoisan. The patterns of population structure revealed highlight the need for regional reference databases that include ethno-linguistic and geographic location information. These markers have important potential for bio-anthropological studies as well as for forensic applications.

GlobalFiler® Express DNA amplification kit in South Africa: Extracting the past from the present

In this study, the GlobalFiler® Express amplification kit was evaluated for forensic use in 541 South African individuals belonging to the Afrikaaner, amaXhosa,11Prefixes ama- and isi- are used within Bantu populations to indicate population group and language respectively. amaZulu,1 Asian Indian and Coloured population groups.Allelic frequencies, genetic diversity parameters and forensic informative metrics were calculated for each population. A total of 301 alleles were observed ranging between 5 and 44.2 repeat units, 43 were rarely observed partial repeats and seven were novel. The combined match probability (CMP) ranged from 2.21×10−26 (Coloured) to 5.21×10−25 (AmaZulu), and the combined power of exclusion (CPE) 0.9999999978 (Afrikaaner) to 0.99999999979 (AmaZulu) respectively. No significant departures from Hardy-Weinberg equilibrium (HWE) were observed after Bonferroni correction.Strong evidence of genetic structure was detected using the coancestry coefficient θ, Analysis of Molecular Variance (AMOVA) and an unsupervised Bayesian clustering method (STRUCTURE). The efficiency of assignment of individuals to population groups was evaluated by applying likelihood ratios with WHICHRUN, and the individual ancestral membership probabilities inferred by STRUCTURE. Likelihood ratios performed the best in the assignment of individuals to population groups. Signs of positive selection were detected for TH01 and D13S317 and purifying/balancing selection for locus SE33. These three loci also displayed the largest informativeness for assignment (In) values.The results of this study supports the use of the GlobalFiler® STR profiling kit for forensic applications in South Africa with the additional capability to predict ethnicity or continental origin of a random sample.

Implementing genotypic AmpFlSTR® Identifiler® Plus profiles to infer population groups

In this study Identifiler® Plus was evaluated in its capacity to infer ancestry by comparing genotypic data garnered for several populations.Implementing pairwise F statistics (Fst) and an unsupervised Bayesian clustering method (STRUCTURE) significant genetic structure was observed between the African, European and Asian population groups. The Khoi-San group Ju\\’hoansi showed the highest differentiation (Fst) with the non-African populations.High correct assignment rates were obtained using likelihood ratios with low error rates, this highlights the significant population structure found. A frequentist method was employed to identify Fst outliers and locus TH01 was shown to be under positive selective pressure.In this study a forensic kit for human identification has been shown to contain important ancestry information which may aid forensics case work if a sufficiently large reference database is established.

Improving the inference of population genetic structure in the presence of related individuals.

It is well known that the presence of related individuals can affect the inference of population genetic structure from molecular data. This has been verified, for example, on the unsupervised Bayesian clustering algorithm implemented in the software STRUCTURE. This methodology assumes, among others, Hardy-Weinberg and linkage equilibrium within subpopulations. The existence of groups of close relatives, such as full-sib families, may prevent these assumptions to be fulfilled, causing the algorithm to work suboptimally. The purpose of this study was to evaluate the effect of the presence of related individuals on a different methodology (implemented in CLUSTER_DIST) for population genetic structure inference. This approach arranges individuals to maximize the genetic distance between groups and does not make Hardy-Weinberg and linkage equilibrium assumptions. We study the robustness of this approach to the presence of close relatives in a sample using simulated scenarios involving combinations of several factors, including the number of subpopulations, the level of differentiation between them, the number, size and type (full or half-sibs) of families in a sample, and the type and number of molecular markers available for clustering analysis. Results indicate that the methodology that maximizes the genetic distance between subpopulations is less influenced by the presence of related individuals than the program STRUCTURE. Therefore, the former can be used, in combination with the program STRUCTURE, to analyse population genetic structure when related individuals are suspected to be present in a sample.

EPA-1233 – Meaningful psychiatric phenotypes emerge from a bayesian taxonomy of eye movements

Use of DSM and ICD nosology to stratify adult psychiatric illnesses poses significant practical challenges for biomarker discovery. Illness categories and subtypes have not been validated by objective tests. The epistemology and language of diagnostic psychiatric tools prevails during 2013–2014 while continuing to say nothing about the biological underpinnings of mental disorders. Assays mapped onto the current nomenclature remain contentious because empirical analyses and clinical intuition are often in conflict, producing’misclassifications’ that are difficult to interpret. DSM-5 revisions are obliged to incorporate empirical psycho-bio-physiological measurements supporting graduated dimensional descriptions of mental health disorders that properly reflect both quanta and spectra of variation extant in the population. Eye movements elicited by simple cognitive tasks are an endophenotypic bridge between the genetics, neuropathology, cognition, and ethno-sociocultural factors of the major neurodevelopmental psychiatric disorders. We have shown that multivariate measures of oculomotor dysfunction are capable of differentiating schizophrenia, bipolar disorder and endogenous major depression cases with exceptional specificity (Benson et al (2012) Eur Psychiatry 27(Supp1):1; Benson et al (2013) Eur Arch Psychiatry Clin Neurosci 263(Supp1): S19). Here we used unsupervised Bayesian clustering to stratify cases and controls using only their eye movement data. Performance measures from smooth pursuit, free-viewing and steady fixation tasks were used. Twelve clusters described a continuum of phenotypes, associated with schizophrenia, bipolar, mixed affective cases, and undiagnosed controls. Composite clusters evinced cases with shared and/or comorbid symptoms and subclinical groups. Eye movement abnormalities through their task-related substrates represent an objective bottom-up approach to diagnosis outside current categorical systems.

The effect of close relatives on unsupervised Bayesian clustering algorithms in population genetic structure analysis

The inference of population genetic structures is essential in many research areas in population genetics, conservation biology and evolutionary biology. Recently, unsupervised Bayesian clustering algorithms have been developed to detect a hidden population structure from genotypic data, assuming among others that individuals taken from the population are unrelated. Under this assumption, markers in a sample taken from a subpopulation can be considered to be in Hardy-Weinberg and linkage equilibrium. However, close relatives might be sampled from the same subpopulation, and consequently, might cause Hardy-Weinberg and linkage disequilibrium and thus bias a population genetic structure analysis. In this study, we used simulated and real data to investigate the impact of close relatives in a sample on Bayesian population structure analysis. We also showed that, when close relatives were identified by a pedigree reconstruction approach and removed, the accuracy of a population genetic structure analysis can be greatly improved. The results indicate that unsupervised Bayesian clustering algorithms cannot be used blindly to detect genetic structure in a sample with closely related individuals. Rather, when closely related individuals are suspected to be frequent in a sample, these individuals should be first identified and removed before conducting a population structure analysis.

Clustering 16S rRNA for OTU prediction: a method of unsupervised Bayesian clustering

Motivation: With the advancements of next-generation sequencing technology, it is now possible to study samples directly obtained from the environment. Particularly, 16S rRNA gene sequences have been frequently used to profile the diversity of organisms in a sample. However, such studies are still taxed to determine both the number of operational taxonomic units (OTUs) and their relative abundance in a sample.Results: To address these challenges, we propose an unsupervised Bayesian clustering method termed Clustering 16S rRNA for OTU Prediction (CROP). CROP can find clusters based on the natural organization of data without setting a hard cut-off threshold (3%/5%) as required by hierarchical clustering methods. By applying our method to several datasets, we demonstrate that CROP is robust against sequencing errors and that it produces more accurate results than conventional hierarchical clustering methods.Availability and Implementation: Source code freely available at the following URL: http://code.google.com/p/crop-tingchenlab/, implemented in C++ and supported on Linux and MS Windows.Contact: tingchen@usc.eduSupplementary information: Supplementary data are available at Bioinformatics online.

Molecular Analysis of Non–Small Cell Lung Cancer Identifies Subsets with Different Sensitivity to Insulin-like Growth Factor I Receptor Inhibition

This study aimed to identify molecular determinants of sensitivity of non-small cell lung cancer (NSCLC) to anti-insulin-like growth factor receptor (IGF-IR) therapy. A total of 216 tumor samples were investigated, of which 165 consisted of retrospective analyses of banked tissue and an additional 51 were from patients enrolled in a phase II study of figitumumab, a monoclonal antibody against IGF-IR, in stage IIIb/IV NSCLC. Biomarkers assessed included IGF-IR, epidermal growth factor receptor, IGF-II, IGF-IIR, insulin receptor substrate 1 (IRS-1), IRS-2, vimentin, and E-cadherin. Subcellular localization of IRS-1 and phosphorylation levels of mitogen-activated protein kinase and Akt1 were also analyzed. IGF-IR was differentially expressed across histologic subtypes (P = 0.04), with highest levels observed in squamous cell tumors. Elevated IGF-IR expression was also observed in a small number of squamous cell tumors responding to chemotherapy combined with figitumumab (P = 0.008). Because no other biomarker/response interaction was observed using classical histologic subtyping, a molecular approach was undertaken to segment NSCLC into mechanism-based subpopulations. Principal component analysis and unsupervised Bayesian clustering identified three NSCLC subsets that resembled the steps of the epithelial to mesenchymal transition: E-cadherin high/IRS-1 low (epithelial-like), E-cadherin intermediate/IRS-1 high (transitional), and E-cadherin low/IRS-1 low (mesenchymal-like). Several markers of the IGF-IR pathway were overexpressed in the transitional subset. Furthermore, a higher response rate to the combination of chemotherapy and figitumumab was observed in transitional tumors (71%) compared with those in the mesenchymal-like subset (32%; P = 0.03). Only one epithelial-like tumor was identified in the phase II study, suggesting that advanced NSCLC has undergone significant dedifferentiation at diagnosis. NSCLC comprises molecular subsets with differential sensitivity to IGF-IR inhibition.

Design of multiple-level hybrid classifier for intrusion detection system using Bayesian clustering and decision trees

With increasing connectivity between computers, the need to keep networks secure progressively becomes more vital. Intrusion detection systems (IDS) have become an essential component of computer security to supplement existing defenses. This paper proposes a multiple-level hybrid classifier, a novel intrusion detection system, which combines the supervised tree classifiers and unsupervised Bayesian clustering to detect intrusions. Performance of this new approach is measured using the KDDCUP99 dataset and is shown to have high detection and low false alarm rates.

Inference of ancestry: constructing hierarchical reference populations and assigning unknown individuals

The ability to infer personal genetic ancestry is being increasingly utilised in certain medical and forensic situations. Herein, the unsupervised Bayesian clustering algorithms structure, is employed to analyse 377 autosomal short tandem repeats typed on 1,056 individuals from the Centre d'Etude du Polymorphisme Humain Human Diversity Panel. Individuals of known geographical origin were hierarchically classified into a framework of increasingly homogeneous clusters to serve as reference populations into which individuals of unknown ancestry can be assigned. The groupings were characterised by the geographical affinities of cluster members and the accuracy of these procedures was verified using several genetic indices. Fine-scale substructure was detectable beyond the broad population level classifications that previously have been explored in this dataset. Metrics indicated that within certain lines, the strongest structuring signals were detected at the leaves of the hierarchy where lineage-specific groupings were identified. The accuracy of unknown assignment was assessed at each level of the hierarchy using a 'leave one out' strategy in which each individual was stripped of cluster membership and then re-assigned using the supervised Bayesian clustering algorithm implemented in GeneClass2. Although most clusters at all levels of resolution experienced highly accurate assignment, a decline was observed in the finer levels due to the mixed membership characteristics of some individuals. The parameters defined by this study allowed for assignment of unknown individuals to genetically defined clusters with measured likelihood. Shared ancestry data can then be inferred for the unknown individual.

Software Sensor Design Using Bayesian Automatic Classification and Back-Propagation Neural Networks

Back-propagation neural networks (BPNN) have attracted attention as an effective method for designing software sensors. A critical issue with BPNN is the danger of extrapolation beyond the parameter space used for the training data. It is therefore important to select the data for model development and test with some care. This also means that there is a need to know when the BPNN model needs to be retrained with new data during use. This paper describes an approach for addressing this issue, which combines an unsupervised clustering method in conjunction with a BPNN model. An unsupervised Bayesian clustering system is used to automatically group the multivariate data into clusters in such a way that data patterns within a class have similar characteristics which distinguish them from other classes. Test data patterns for the BPNN model are selected from each class and when new data are available, the clustering system is employed to check if they are beyond the parameter space of the previous training da...