Search

Summary Calcium, magnesium and phosphorus are essential electrolytes involved in a large number of biological processes. Imbalance of these minerals in blood may indicate clinically relevant conditions and are important in inferring acute or chronic pathologies in humans and animals. In this work, we carried out a genome-wide association study (GWAS) for the level of these three electrolytes in the serum of 843 performance-tested Italian Large White pigs. All pigs were genotyped with the Illumina PorcineSNP60 BeadChip, and GWAS was carried out using genome-wide efficient mixed-model association. For the level of Ca2+, eight single nucleotide polymorphisms (SNPs) were significant, considering a false discovery rate (FDR)

Metabolomics has opened new possibilities to investigate metabolic differences among animals. In this study, we applied a targeted metabolomic approach to deconstruct the pig sex metabolome as defined by castrated males and entire gilts. Plasma from 545 performance-tested Italian Large White pigs (172 castrated males and 373 females) sampled at about 160 kg live weight were analyzed for 186 metabolites using the Biocrates AbsoluteIDQ p180 Kit. After filtering, 132 metabolites (20 AA, 11 biogenic amines, 1 hexose, 13 acylcarnitines, 11 sphingomyelins, 67 phosphatidylcholines, and 9 lysophosphatidylcholines) were retained for further analyses. The multivariate approach of the sparse partial least squares discriminant analysis was applied, together with a specifically designed statistical pipeline, that included a permutation test and a 10 cross-fold validation procedure that produced stability and effect size statistics for each metabolite. Using this approach, we identified 85 biomarkers (with metabolites from all analyzed chemical families) that contributed to the differences between the 2 groups of pigs ( < 0.05 at the stability statistic test). All acylcarnitines and almost all biogenic amines were higher in castrated males than in gilts. Metabolites involved in tryptophan catabolism had the largest differences (i.e., delta = 20% for serotonin) between castrated males (higher) and gilts (lower). The level of several AA (Ala, Arg, Gly, His, Lys, Ser, Thr, and Trp) was higher in gilts (delta was from approximately 1.0 to approximately 4.8%) whereas products of AA catabolism (taurine, 2-aminoadipic acid, and methionine sulfoxide) were higher in castrated males (delta was approximately 5.0-6.0%), suggesting a metabolic shift in castrated males toward energy storage and lipid production. Similar general patterns were observed for most sphingomyelins, phosphatidylcholines, and lysophosphatidylcholines. Metabolomic pathway analysis and pathway enrichment identified several differences between the 2 sexes. This metabolomic overview opened new clues on the biochemical mechanisms underlying sexual dimorphism that, on one hand, might explain differences in terms of economic traits between castrated male pigs and entire gilts and, on the other hand, could strengthen the pig as a model to define metabolic mechanisms related to fat deposition.

We genotyped 58 single nucleotide polymorphisms (SNPs) in 25 candidate genes in about 800 Italian Holstein sires. Fifty-six (minor allele frequency >0.02) were used to evaluate their association with single traits: milk yield (MY), milk fat yield (FY), milk protein yield (PY), milk fat percentage (FP), milk protein percentage (PP), milk somatic cell count (MSCC); and complex indexes: longevity, fertility and productivity-functionality type (PFT), using deregressed proofs, after adjustment for familial relatedness. Thirty-two SNPs were significantly associated (proportion of false positives

Average daily gain is an important target trait in pig breeding programs. In this study we performed a genomewide association study for ADG in Italian Large White pigs using a selective genotyping approach. Two extreme and divergent groups of Italian Large White pigs (number 190 + 190) were selected among a population of about 10,000 performance tested gilts (EBV for ADG in the 2 groups were -30 ± 14 g and 81 ± 12 g, respectively) and genotyped with the Illumina PorcineSNP60 BeadChip. Association analysis was performed treating the pigs of the 2 extreme groups as cases and controls after correction for family-based stratification. A total of 127 SNP resulted significantly associated with ADG (P nominal value [P(raw)] < 2.0 × 10(-7), P < 0.01 Bonferroni corrected [P(Bonferroni)] < 0.01, false discovery rate < 7.76 × 10(-5)). Another 102 SNP were suggestively associated with the target trait (P(raw) between 2.0 × 10(-7) and 2.02 × 10(-6), P(Bonferroni) < 0.10, false discovery rate < 4.19 × 10(-4)). These SNP were located on all autosomes and on porcine chromosome (SSC) X. The largest number of SNP within this list was on SSC5 (n = 42), SSC7 (34), SSC6 (30), SSC4 (23), and SSC16 (16). These chromosomes were richer in significant or suggestively significant markers than expected (P < 0.001). A quite high number of these SNP (n = 23) were associated with backfat thickness in a previous genomewide association study performed in the same pig population, confirming the negative correlation between the 2 traits. Two or more SNP targeted the same gene: IGSF3 and HS2ST1 (SSC4), OTOGL (SSC5), FTO region (SSC6), and MYLK4 and MCUR1 (SSC7). Other regions that were associated with ADG in previous candidate gene studies (e.g., MC4R on SSC1, IGF2 and LDHA on SSC2, MUC4 on SSC13) 1) included markers with P(raw) < 0.01 that, however, did not pass the stringent threshold of significance adopted in this study or 2) could not be tested because not assigned to the Sscrofa10.2 genome version. Functional annotation of the significant regions using Gene Ontology suggested that many and complex processes at different levels are involved in affecting ADG, indicating the complexity of the genetic factors controlling this ultimate phenotype. The obtained results may contribute to understand the genetic mechanisms determining ADG that could open new perspectives to improve selection efficiency in this breed.

The problem of clustering probability density functions is emerging in different scientific domains. The methods proposed for clustering probability density functions are mainly focused on univariate settings and are based on heuristic clustering solutions. New aspects of the problem associated with the multivariate setting and a model-based perspective are investigated. The novel approach relies on a hierarchical mixture modeling of the data. The method is introduced in the univariate context and then extended to multivariate densities by means of a factorial model performing dimension reduction. Model fitting is carried out using an EM-algorithm. The proposed method is illustrated through simulated experiments and applied to two real data sets in order to compare its performance with alternative clustering strategies.

Melanocortin 4 receptor (MC4R) plays a key role in controlling energy homeostasis. Several studies have already reported effects on production traits of polymorphisms identified in the porcine MC4R gene. In this study we analysed data on 6 MC4R polymorphisms (c.-780C>G; c.-135C>T; c.175C>T; c.707G>A or p.Arg236His; and c.892G>A or p.Asp298Asn; c.*430A>T) genotyped from (1) two groups of Italian Large White pigs (280+280 animals) with extreme estimated breeding values (EBVs) for back fat thickness (BFT), selected among a performance tested population of about 12,000 pigs, and from (2) 19 Italian Duroc pigs. Two haplotypes, differentiated by the c.892G>A, were identified in the Duroc populations. Four haplotypes were identified in the Italian Large White population, one of which (haplotype 4) was identified for the first time in this study. Single marker and haplotype association analyses for BFT were obtained by comparing allele and haplotype frequency differences from the two extreme tails using χ2 and Cochran-Armitage trend tests. Results confirmed the effects of the c.892G>A single nucleotide polymorphism (SNP) on BFT, as also defined by different distributions in the two tails of haplotypes carrying the alternative nucleotides at this polymorphic site (P

Combining different approaches (resequencing of portions of 54 obesity candidate genes, literature mining for pig markers associated with fat deposition or related traits in 77 genes, and in silico mining of porcine expressed sequence tags and other sequences available in databases), we identified and analyzed 736 SNP within candidate genes to identify markers associated with back fat thickness (BFT) in Italian Large White sows. Animals were chosen using a selective genotyping approach according to their EBV for BFT (276 with most negative and 279 with most positive EBV) within a population of ≈ 12,000 pigs. Association analysis between the SNP and BFT has been carried out using the MAX test proposed for case-control studies. The designed assays were successful for 656 SNP: 370 were excluded (low call rate or minor allele frequency A polymorphism (P(nominal) G polymorphism (P(nominal) = 8.0E-05). The third top SNP (P(nominal) = 6.2E-04) was the intronic TBC1D1 g.219G>A polymorphic site, in agreement with our previous results obtained in an independent study. The list of significant markers also included SNP in additional genes (ABHD16A, ABHD5, ACP2, ALMS1, APOA2, ATP1A2, CALR, COL14A1, CTSF, DARS, DECR1, ENPP1, ESR1, GH1, GHRL, GNMT, IKBKB, JAK3, MTTP, NFKBIA, NT5E, PLAT, PPARG, PPP2R5D, PRLR, RRAGD, RFC2, SDHD, SERPINF1, UBE2H, VCAM1, and WAT). Functional relationships between genes were obtained using the Ingenuity Pathway Analysis (IPA) Knowledge Base. The top scoring pathway included 19 genes with a P(nominal) < 0.1, 2 of which (IKBKB and NFKBIA) are involved in the hypothalamic IKKβ/NFκB program that could represent a key axis to affect fat deposition traits in pigs. These results represent a starting point to plan marker-assisted selection in Italian Large White nuclei for BFT. Because of similarities between humans and pigs, this study might also provide useful clues to investigate genetic factors affecting human obesity.

A pilot study using Affymetrix Gene Chip(R) Porcine Genome Arrays was set up to evaluate the impact of time lags from death on gene expression profiling of porcine skeletal muscle at four post mortem times (up to 24 h) during the routine processing of fresh thighs. All post chip parameters and data analyses (Average background, Scale Factors, Percent Presence, 3'/5' ratios of beta-actin and glyceraldehyde-3-phosphase dehydrogenase control genes, RNA degradation diagnostics, principal component analysis, hierarchical clustering, mixed regression models with time effects) did not show any effect of post mortem time. Therefore, microarray data obtained from muscle specimens collected in a processing plan over a quite long period have the potential to identify treatments or pre mortem conditions without any potential bias derived from subtle RNA degradation. These results open new perspectives to develop and analyse gene expression biomarkers for pig production and product authentication.

Recently, different mixture models have been proposed for multilevel data, generally requiring the local independence assumption. In this work, this assumption is relaxed by allowing each mixture component at the lower level of the hierarchical structure to be modeled according to a multivariate Gaussian distribution with a non-diagonal covariance matrix. For high-dimensional problems, this solution can lead to highly parameterized models. In this proposal, the trade-off between model parsimony and flexibility is governed by assuming a latent factor generative model.

In the general classification context the recourse to the so-called Bayes decision rule requires to estimate the class conditional probability density functions. In this paper we propose a mixture model for the observed variables which is derived by assuming that the data have been generated by an independent factor model. Independent factor analysis is in fact a generative latent variable model whose structure closely resembles the one of ordinary factor model but it assumes that the latent variables are mutually independent and not necessarily Gaussian. The method therefore provides a dimension reduction together with a semiparametric estimate of the class conditional probability density functions. This density approximation is plugged into the classic Bayes rule and its performance is evaluated both on real and simulated data.

In this study, we investigated at the genome-wide level if 20 years of artificial directional selection based on boar genetic evaluation obtained with a classical BLUP animal model shaped the genome of the Italian Large White pig breed. The most influential boars of this breed (n = 192), born from 1992 (the beginning of the selection program of this breed) to 2012, with an estimated breeding value reliability of >0.85, were genotyped with the Illumina Porcine SNP60 BeadChip. After grouping the boars in eight classes according to their year of birth, filtered single nucleotide polymorphisms (SNPs) were used to evaluate the effects of time on genotype frequency changes using multinomial logistic regression models. Of these markers, 493 had a PBonferroni

In this paper, the potentialities of transvariation (Gini, 1959) in measuring the separation between two groups of multivariate observations are explored. With this aim, a modified version of Gini’s notion of multidimensional transvariation is proposed. According to Gini (1959), two groups G1 and G2 are said to transvary on the k-dimensional variable X = (X1,...,Xh,...,Xk) if there exists at least one pair of units, belonging to different groups, such that for h = 1,...,k the sign of the difference between their Xh values is opposite to that of m1h −m2h, where m1h and m2h are the corresponding group mean values of Xh. We introduce a modification that allows us to derive a measure of group separation, which can be profitably used in discriminating between two groups. The performance of the measure is tested through simulation experiments. The results show that the proposed measure is not sensitive to distributional assumptions and highlight its robustness against outliers.

Summary The genetic identification of the population of origin of individuals, including animals, has several practical applications in forensics, evolution, conservation genetics, breeding and authentication of animal products. Commercial high-density single nucleotide polymorphism (SNP) genotyping tools that have been recently developed in many species provide information from a large number of polymorphic sites that can be used to identify population-/breed-informative markers. In this study, starting from Illumina BovineSNP50 v1 BeadChip array genotyping data available from 3711 cattle of four breeds (2091 Italian Holstein, 738 Italian Brown, 475 Italian Simmental and 407 Marchigiana), principal component analysis (PCA) and random forests (RFs) were combined to identify informative SNP panels useful for cattle breed identification. From a PCA preselected list of 580 SNPs, RFs were computed using ranking methods (Mean Decrease in the Gini Index and Mean Accuracy Decrease) to identify the most informative 48 and 96 SNPs for breed assignment. The out-of-bag (OOB) error rate for both ranking methods and SNP densities ranged from 0.0 to 0.1% in the reference population. Application of this approach in a test population (10% of individuals pre-extracted from the whole data set) achieved 100% of correct assignment with both classifiers. Linkage disequilibrium between selected SNPs was relevant (r2 > 0.6) only in few pairs of markers indicating that most of the selected SNPs captured different fractions of variance. Several informative SNPs were in genes/QTL regions that affect or are associated with phenotypes or production traits that might differentiate the investigated breeds. The combination of PCA and RF to perform SNP selection and breed assignment can be easily implemented and is able to identify subsets of informative SNPs useful for population assignment starting from a large number of markers derived by high-throughput genotyping platforms.

SummaryIn this study, we investigated whether a selection programme based onboar genetic evaluation obtained with a classical BLUP animal model canchange allele frequencies in a pig population. All Italian Large White boarsborn from 1992 to 2012 with estimated breeding value reliability >0.85(n = 200) were selected among all boars of this breed. Boars were geno-typed with markers in major genes (IGF2 intron3-g.3072G>A, MC4Rp.D298N, VRTN PRE1 insertion, PRKAG3 p.I199V and FTO g.276T>G).Genotyping data were analysed grouping boars in eight classes accordingto their year of birth. To evaluate the influence of time on allele frequen-cies of the genotyped markers, multinomial logistic regression modelswere computed. Four of five polymorphic sites (IGF2, MC4R, VRTN andFTO) showed significant (p < 0.01) changes in allele frequencies over timedue to a progressive and continuous increase of one allele (associated withhigher lean meat content, higher average daily gain and favourable feed:gain ratio) and, consequently, decrease of the other one, following thedirectional selection of the selection programme of this pig breed. The ret-rospective analysis that was carried out in Italian Large White boars sug-gests that selection based on methodologies assuming the infinitesimalmodel is able to modify in a quite short period of time allele frequencies inmajor genes, increasing the frequency of alleles explaining a relevant(non-infinitesimal) fraction of the overall genetic variability for produc-tion traits.IntroductionSelection in livestock populations has been mainlybased on methodologies assuming the infinitesimalmodel that describes quantitative traits as determinedby an infinite number of genes, each of them with avery small effect. By this approach, genetic improve-ment is not expected to significantly change allelefrequencies at any particular locus in a population(Barton & Keightley 2002). However, experimentsdissecting genetic variability in livestock genomeshave substantially challenged the assumptions of theinfinitesimal model showing that some loci explain arelevant part of the genetic variance of quantitativetraits under selection. These loci have been termedquantitative trait loci (QTL) or major genes when thecausative genes and mutations are known and theireffects are particularly relevant. Therefore, directional

The aim of this study was to identify copy number variants (CNVs) in Italian Large White pigs and test them for association with back fat thickness (BFT). Within a population of 12 000 performance-tested pigs, two groups of animals with extreme and divergent BFT estimated breeding values (EBVs; 147 with negative and 150 with positive EBVs) were genotyped with the Illumina Porcine SNP60 BeadChip. CNVs were detected with PENNCNV software. We identified a total of 4146 CNV events in 170 copy number variation regions (CNVRs) located on 15 porcine autosomes. Validation of detected CNVRs was carried out (i) by comparing CNVRs already detected by other studies and (ii) by semiquantitative fluorescent multiplex (SQFM) PCR of a few CNVRs. Most of CNVRs detected in Italian Large White pigs (71.2%) were already reported in other pig breeds/populations, and 82.1% of the CNV events detected by PENNCNV were confirmed by SQFM PCR. For each CNVR, we compared the occurrence of CNV events between the pigs of the high and low BFT EBV tails. Sixteen regions showed significance at P < 0.10, and seven were significant at P < 0.05 but were not significant after Bonferroni correction (Fisher's exact test). These results indicated that CNVs could explain a limited fraction of the genetic variability of fat deposition in Italian Large White pigs. However, it was interesting to note that one of these CNVRs encompassed the ZPLD1 gene. In humans, a rare CNV event including this gene is associated with obesity. Studies identifying CNVs in pigs could assist in elucidating the genetic mechanisms underlying human obesity.

Background Carcass fatness is an important trait in most pig breeding programs. Following market requests, breeding plans for fresh pork consumption are usually designed to reduce carcass fat content and increase lean meat deposition. However, the Italian pig industry is mainly devoted to the production of Protected Designation of Origin dry cured hams: pigs are slaughtered at around 160 kg of live weight and the breeding goal aims at maintaining fat coverage, measured as backfat thickness to avoid excessive desiccation of the hams. This objective has shaped the genetic pool of Italian heavy pig breeds for a few decades. In this study we applied a selective genotyping approach within a population of ~ 12,000 performance tested Italian Large White pigs. Within this population, we selectively genotyped 304 pigs with extreme and divergent backfat thickness estimated breeding value by the Illumina PorcineSNP60 BeadChip and performed a genome wide association study to identify loci associated to this trait. Results We identified 4 single nucleotide polymorphisms with P≤5.0E-07 and additional 119 ones with 5.0E-07 Conclusions Further investigations are needed to evaluate the effects of the identified single nucleotide polymorphisms associated with backfat thickness on other traits as a pre-requisite for practical applications in breeding programs. Reported results could improve our understanding of the biology of fat metabolism and deposition that could also be relevant for other mammalian species including humans, confirming the role of neuronal genes on obesity.

The Forward Search (FS) represents a useful tool for clustering data that include outlying observations, because it provides a robust clustering method in conjunction with graphical tools for outlier identification. In this paper, we show that recasting FS clustering in the framework of normal mixture models can introduce some improvements: the problem of choosing a metric for clustering is avoided; membership degree is assessed by posterior probability; a testing procedure for outlier detection can be devised.

Complex data such as those where each statistical unit under study is described not by a single observation (or vector variable), but by a unit-specific sample of several or even many observations, are becoming more and more popular. Reducing these sample data by summary statistics, like the average or the median, implies that most inherent information (about variability, skewness or multi-modality) gets lost. Full information is preserved only if each unit is described by a whole distribution. This new kind of data, a.k.a. "distribution-valued data", require the development of adequate statistical methods. This paper presents a method to group a set of probability density functions (pdfs) into homogeneous clusters, provided that the pdfs have to be estimated nonparametrically from the unit-specific data. Since elements belonging to the same cluster are naturally thought of as samples from the same probability model, the idea is to tackle the clustering problem by defining and estimating a proper mixture model on the space of pdfs. The issue of model building is challenging here because of the infinite-dimensionality and the non-Euclidean geometry of the domain space. By adopting a wavelet-based representation for the elements in the space, the task is accomplished by using mixture models for hyper-spherical data. The proposed solution is illustrated through a simulation experiment and on two real data sets.

Thirty-nine glial tumours (28 glioblastomas (GB) and 11 low-grade gliomas) were investigated with DNA microarrays to reveal a possible specific gene expression profile. Unsupervised classification through hierarchical cluster analysis identified two groups of tumours, the first composed of low-grade gliomas and the second mainly composed of GB. Nine genes were identified as most informative: seven were over-expressed in low-grade gliomas and under-expressed in GB; on the contrary, two genes, insulin-like growth factor binding protein 2 (IGFBP-2) and cell division cycle 20 homologue (CDC20), were over-expressed in GB and under-expressed in low-grade tumours. This same genetic profile was confirmed by reverse transcriptase polymerase chain reaction. Immunohistochemistry for IGFBP-2 was positive in 88.8% of the cases of GB and in only one low-grade glioma, whilst CDC20 immunostained 74.1% of the cases of GB and none low-grade glioma. This was confirmed in an additional series of cases studied with immunohistochemistry only. In conclusion, over-expression of mRNA levels of IGFBP-2 and CDC20 is highly related to GB, IGFBP-2 and CDC-20 gene and protein expressions are strongly correlated, and IGFBP-2 and CDC20 immunopositivity can be useful for the identification of GB in small biopsies.

This paper proposes a method based on CLV (Clustering around Latent Variables; [Vigneau, E., & Qannari, E. M. (2003). Clustering of variables around latent components. Communications in Statistics Simulation and Computation, 12 (4), 1131–1150] for identifying groups of consumers in L-shaped data structures. This kind of data structure is very common in consumer studies, when a panel of consumers is asked to assess overall liking of a certain number of products and the preference scores are then arranged in a two-way table, Y. External information both on products (physical–chemical characterisation or sensory profile) and consumers (socio-demographic background, purchasing behaviour or consumption habits) may be available respectively in a row-descriptor matrix, X and a column-descriptor matrix, Z . The aim is to automatically provide consumer segmentation, with all three matrices playing an active role in classification, obtaining groups which are as homogeneous as possible from every point of view: preferences, products and consumer characteristics. The proposed clustering method is compared with the procedure adopted in [Esposito Vinzi, V., Guinot, C., & Squillacciotti, S. (2007). Two-step PLS regression for L-structured data: An application in the cosmetics industry. Statistical Methods and Applications, 16 (2), 263–278], in which double PLS regression is used to capture all the information contained in the L-structure and consumer classification is then carried out on the columns of the final component matrix. Furthermore, the two approaches are illustrated using data from a preference study on juices based on berry fruits. The hedonic ratings of 72 consumers for 25 fruit juice mixes have been explained with regard to the product’s chemical characterisation and socio-demographic information, purchasing behaviour and the consumption habits of consumers.

Forward search (FS) methods have been shown to be usefully employed for detecting multiple outliers in continuous multivariate data (Hadi, (1994); Atkinson et al., (2004)). Starting from an outlier-free subset of observations, they iteratively enlarge this good subset using Mahalanobis distances based only on the good observations. In this paper, an alternative formulation of the FS paradigm is presented, that takes a mixture of K > 1 normal components as a null model. The proposal is developed according to both the graphical and the inferential approach to FS-based outlier detection. The performance of the method is shown on an illustrative example and evaluated on a simulation experiment in the multiple cluster setting.

In this article, we present a strategy for producing low-dimensional projections that maximally separate the classes in Gaussian Mixture Model classification. The most revealing linear manifolds are those along which the classes are maximally separable. Here we consider a particular probability product kernel as a measure of similarity or affinity between the class-conditional distributions. It takes an appealing closed analytical form in the case of Gaussian mixture components. The performance of the proposed strategy has been evaluated on real data.

Gaussian mixture models (GMM) are commonly employed in nonparametric supervised classification. In high-dimensional problems it is often the case that information relevant to the separation of the classes is contained in a few directions. A GMM fitting procedure oriented to supervised classification is proposed, with the aim of reducing the number of free parameters. It resorts to projection pursuit as a dimension reduction method and combines it with GM modelling of class-conditional densities. In its derivation, issues regarding the forward and backward projection pursuit algorithms are discussed. The proposed procedure avoids the “curse of dimensionality”, is able to model structure in subspaces and regularizes the classification model. Its performance is illustrated on a simulation experiment and on a real data set, in comparison with other GMM-based classification methods.

Large histologic sections (LHSs) are increasingly used in the study of normal and neoplastic breast tissue. LHSs allow the direct visualization of a large part of the breast glandular tree. Accordingly, LHSs have shown that in situ and invasive lobular carcinoma is a multilobar (and hence multifocal) neoplastic lesion in more than 50% of the cases, and that poorly differentiated duct carcinoma in situ (DCIS grade 3) is frequently unifocal, whereas it is often multifocal when the in situ lesion is a well-differentiated type (DCIS grade 1). Forty-five mastectomies were studied with large sections. Mastectomies were performed when quadrantectomy did not guarantee radical excision of the tumor with adequate cosmesis because of the large size of the lesion or because the neoplastic lesion was located below the nipple. Excluded were cases of lobular neoplasia or invasive lobular carcinoma, because they were reported separately, and cases of mastectomies performed for sarcoma or recurrent phyllodes tumor. All cases had undergone a preoperative diagnostic procedure (fine needle aspiration), and the relative positive material was reviewed. All 45 cases showed in situ duct carcinoma and 37 showed evidence of invasive duct carcinoma. Forty-two cases of DCIS were multifocal, whereas only 4 invasive duct carcinoma were shown as multifocal. When DCIS lesions were subdivided into 3 grades, no statistical significance was seen among the 3 groups of DCIS in regard to multifocality. Nevertheless, DCIS grade 1 was a widespread condition involving more than one lobe and quadrant, whereas DCIS grades 2 and 3 appeared more localized. DCIS grade 1 was more similar to that previously observed in lobular in situ neoplasia/lobular in situ carcinoma. In 66.6% of the cases, DCIS foci were found within the invasive areas, indicating a more than fortuitous occurrence (2-sided P=.0357).

In this paper we propose a model based density estimation method which is rooted in Independent Factor Analysis (IFA). IFA is in fact a generative latent variable model, whose structure closely resembles the one of an ordinary factor model but which assumes that the latent variables are mutually independent and distributed according to Gaussian mixtures. From these assumptions, the possibility of modelling the observed data density as a mixture of Gaussian distributions too derives. The number of free parameters is controlled through the dimension of the latent factor space. The model is proved to be a special case of mixture of factor analyzers which is less parameterized than the original proposal by McLachlan and Peel (2000). We illustrate the use of IFA density estimation for supervised classification both on real and simulated data.

In this paper the problem of cell classification using gene expression data is addressed. One of the main features of this kind of data is the very large number of variables (genes), relative to the number of observations (cells). This condition makes most of the standard statistical methods for classification difficult to employ. The proposed solution consists of building classification rules on subsets of genes showing a behavior across the cells that differs most from that of all the other ones. This variable selection procedure is based on suitable linear transformations of the observed data: a strategy resorting to independent component analysis is explored. Our proposal is compared with the nearest shrunken centroid method (Tibshirani et al. (2002)) on three publicly available data sets.

One of the most important and broadly accepted solutions for the centre selection problem in radial basis function networks is due to Moody and Darken (1989), who propose to cluster the input vectors via the k-means clustering method. In this paper, an alternative robust solution, based on the spatial median, is proposed and compared also to the k-medoids method (Kaufman and Rousseeuw, 1990). Some simulation studies in the context of classification problems show that, when outlying data are present, the solution we propose improves the network performance and allows to define a more parsimonious network.

Starting from the results obtained by recursive partitioning methods, radial basis function networks for longitudinal data are derived. The aim of this work is to show how the joint use of the two methods allows to overcome some drawbacks that they show when they are used separately. More precisely, this strategy allows not only to obtain a smooth nonparametric estimate of the regression surface, but also to automatically determine the model complexity and to perform a covariate selection. The performances of the proposed strategy are evaluated on simulated data sets.

Radial basis function (RBF) neural networks can be used for a wide range of applications as they can be regarded as universal approximators and their training is faster than that of multilayer perceptrons. A more general version of these neural networks are referred to as elliptical basis function (EBF) networks. In this paper a robust method for EBF parameter estimation is proposed, based on hyperellipsoidal clustering and on multivariate sign and rank concepts. A simulation study on a classification problem has shown that this method represents a valid learning scheme, particularly in presence of outlying data.

In this paper a nonparametric method for discriminant analysis is proposed, based on a group separation oriented version of projection pursuit density estimation. Each population is separated in turn from the remaining ones, considered as a whole, by approximating the boundary between them through the composition of some informative directions, chosen according to an appropriate discrimination criterion. A coherent allocation rule is proposed, too. Simulation studies have shown that this method represents a valid solution for problems when the parametric approaches are not flexible enough and sample sizes are too small to use classical nonparametric methods.

In this paper we derive a two group-linear discriminant function (LDF) by minimizing Gini’s transvariation probability. This solution is a special case of projection pursuit methods and improves the performances of Fisher’s LDF when the conditions which guarantee its optimality do not hold.

Fatness in pigs is a complex trait for which a large number of genes are expected to be involved. Genetics of human obesity could take advantages from genetic information coming from the pig and vice versa. To these aims, a comprehensive candidate gene approach could be helpful. We catalogued all genes affecting fatness on both species, and identified in silico and by resequencing porcine SNPs on a large number of candidate genes. In addition, we applied a selective genotyping approach to identify markers associated with fat deposition in pigs. This approach was tested genotyping the IGF2 intron3-g.3072G>A mutation and novel markers in the PCSK1 and TBC1D1 genes. Polymorphisms in these genes resulted associated with back fat thickness in Italian Large White pigs.

In pigs, many production traits are known to vary among breeds or lines. These traits can be considered end phenotypes or external traits as they are the final results of complex biological interactions and processes whose fine biological mechanisms are still largely unknown. This study was designed to compare plasma and serum metabolomic profiles between animals of two heavy pig breeds (12 Italian Large White and 12 Italian Duroc), testing indirectly the hypothesis that different genetic backgrounds might be the determining factors of differences observed on the level of metabolites in the analyzed biofluids between breeds. We used a targeted metabolomic approach based on mass spectrometric detection of about 180 metabolites and applied a statistical validation pipeline to identify differences in the metabolomic profiles of the two heavy pig breeds. Blood samples were collected after jugulation at the slaughterhouse and prepared for metabolomics analysis that was carried out using the Biocrates AbsoluteIDQ p180 Kit, covering five different biochemical classes: glycerophospholipids, amino acids, biogenic amines, hexoses and acylcarnitines. A statistical pipeline that included the selection of the most relevant metabolites differentiating the two breeds by sparse Partial Least Squares Discriminant Analysis (sPLS-DA) was coupled with a stability test and significance test determined with leave one out and permutation procedures. sPLS-DA plots clearly separated the pigs of the two investigated breeds. A few metabolites (a total of five metabolites considering the two biofluids) involved in key metabolic pathways largely contributed to these differences between breeds. In particular, a higher level of the sphingomyelins SM (OH) C14:1 (both in plasma and serum), SM (OH) C16:1 (in serum) and SM C16:0 (in serum) were observed in Italian Duroc than in Italian Large White pigs and the inverse was for the biogenic amine kynurenine (in plasma). The level of another biogenic amine (acetylornithine) was higher in Italian Large White than in Italian Duroc pigs in both analysed biofluids. These results provided biomarkers that could be important to understand the biological differences between these two heavy pig breeds. In particular, according to the functional role played by sphingomyelins in obesity-induced inflammatory responses, it could be possible to speculate that a higher level of sphingomyelins in Italian Duroc might be related to the higher interrmuscular fat deposition of this breed compared with the Italian Large White. Additional studies will be needed to evaluate the relevance of these biomarkers for practical applications in pig breeding and nutrition.