Università Cattolica del Sacro Cuore


Mercoledì 15 Novembre 2017 alle ore 14,00

Path weights, networked partial correlations and their application to the analysis of genetic interactions

Prof. Alberto ROVERATO, Università di Bologna

ABSTRACT: Genetic interactions confer robustness on cells in response to genetic perturbations. This often occurs through molecular buffering mechanisms that can be predicted using, among other features, the degree of coexpression between genes, commonly estimated through marginal measures of association such as Pearson or Spearman correlation coefficients. However, marginal correlations are sensitive to indirect effects and often partial correlations are used instead. Yet, partial correlations convey no information about the (linear) influence of the coexpressed genes on the entire multivariate system, which may be crucial to discriminate functional associations from genetic interactions. To address these two shortcomings, here we propose to use the edge weight derived from the covariance decomposition over the paths of the associated gene network. We call this new quantity the networked partial correlation and use it to analyse genetic interactions in yeast.

More concretely, in its well-characterized leucine biosynthesis pathway and on a previously published data set of genome-wide quantitative genetic interaction profiles. In both cases, networked partial correlations substantially improve the identification of genetic interactions over classical coexpression measures. This talk is based on a joint work with Robert Castelo, University Pompeu Fabra, Spain.

Martedì 13 Giugno 2017 alle ore 11,30

Forensic Statistics. A General View

Prof. B. Vittorio FROSINI, Emerito Università Cattolica del Sacro Cuore di Milano

ABSTRACT: Introduction to the Proceedings of a meeting on Forensic Statistics, published by the Italian Association for Applied Statistics (Statistica Applicata – Italian Journal of Applied Statistics, vol. 27, N. 2 and N. 3).A general topic under discussion is devoted to Bayes’ formula (or theorem): it is a fundamental device which allows the computation of probabilities of possible causes, given related information about certain effects. Strictly connected to this argument is the «fallacy of the transposed conditional», which usually implies an unwarranted assumption of guilt. These introductory topics are followed by a discussion of some main features of two causes célèbres, those of Alfred Dreyfus in France (whose law-suit was heavily burdened with many logical errors and fallacies) and of Nicola Sacco and Bartolomeo Vanzetti in U.S.A. (by making use of the extensive and keen examination, made by J.B. Kadane and D.A. Schum, of all the material drawn from the two relative suits). A discussion follows about the role and significance in trials of so-called naked statistics, which most (but not all) experts exclude from applications in trials. Then two interesting cases are presented, related to the Australian case named by Chedzey, and the one related to the English case named by Sally Clark. The final discussion is devoted to two kinds of cases which require a judgment about groups of people; the statistical methods usually applied in these cases pertain to the so-called tests of significance.

Key Words: Bayes’ theorem; Fallacy of the transposed conditional; Naked statistics; Tests of significance.

Giovedì 4 Maggio 2017 alle ore 11,30

Bayesian comparison of classifiers and Learning Bayesian networks with bounded treewidth

Dott. Giorgio CORANI, Università della Svizzera Italiana

ABSTRACT: Usually one compares the accuracy of two competing classifiers via null hypothesis significance tests (nhst). Yet the nhst tests suffer from important shortcomings, which can be overcome by switching to Bayesian hypothesis testing. We discuss a Bayesian hierarchical model which jointly analyzes the cross-validation results obtained by two classifiers on multiple data sets.
Most structural learning approaches for Bayesian networks  do not bound the treewidth k of the DAG. Yet this is very important, since the time-complexity of inference grows exponentially with k.
We present algorithms for learning bounded treewidth Bayesian networks, which scale up to thousands of variables.

Venerdì 21 Aprile 2017 alle ore 11,30

Bayesian approaches for complex networks

Dott. Francesco STINGO, Università degli Studi di Firenze

ABSTRACT: Multi-dimensional  data  constituted  by  measurements  along  multiple  axes  have emerged  across  many  scientific  areas  such  as  genomics  and cancer  surveillance. Traditional multivariate  approaches  are  unsuitable  for  such  highly  structured  data due to inefficiency, loss of power and lack of interpretability. I will illustrate a novel class  of  multi-dimensional  graphical  models  that  includes  both  directed  and undirected graphs as well as arbitrary combinations of these.

Venerdì 10 Marzo 2017 alle ore 11.30

Experimental Design for modeling the flow of particles

Prof. Jesus LOPEZ FIDALGO, Universidad de Castilla-La Mancha, Espana

ABSTRACT: Materials in granular form are widely used, having a great importance in the chemical, food, agricultural and pharmaceutical industries. During the discharge of a two-dimensional silo, the flow of grains through an orifice is arrested if the size of the outlet is not large enough. In the outpouring of grains, jamming occurs due to the formation of an arch at the outlet. After a jam, an input of energy (blowing, shaking or tapping) is necessary to break the blocking arch and restart the flow. Then, the grains fall until a new arch is formed. Several models have been proposed to explain this process. In this talk, optimal experimental designs will be given for some models proposed by Jandaetal. (2008) and To (2005). Design for discriminating between these models will also be provided.

Martedi 14 Febbraio 2017 alle ore 11.00

Random Fields Evolving Temporally over the sphere of Planet Earth

Prof. Emilio PORCU, Department of Mathematics, University Federico Santa Maria, Valparaiso, Chile

ABSTACT: We propose stationary covariance functions for processes that evolve temporally over a sphere, as well as cross-covariance functions for multivariate random fields defined over a sphere. For such processes, the great circle distance is the natural metric that should be used to describe spatial dependence. Given the mathematical difficulties for the construction of covariance functions for processes defined over spheres cross time, approximations of the state of nature have been proposed in the literature by using the Euclidean (based on map projections) and the chordal distances. We present several methods of construction based on the great circle distance and provide closed-form expressions for both spatio-temporal and multivariate cases. A simulation study assesses the discrepancy between the great circle distance, chordal distance, and Euclidean distance based on a map projection both in terms of estimation and prediction in a space-time and a bivariate spatial setting, where the space is in this case the Earth. We revisit the analysis of Total Ozone Mapping Spectrometer (TOMS) data and investigate differences in terms of estimation and prediction between the aforementioned distance-based approaches. Both simulation and real data highlight sensible differences in terms of estimation of the spatial scale parameter. As far as prediction is concerned, the differences can be appreciated only when the interpoint distances are large, as demonstrated by an illustrative example.

January 23, 24, 25; 2017

Short course

Understanding Bayesian Networks with Examples in R

Dott. Marco SCUTARI, Department of Statistics, University of Oxford, UK

The purpose of this short course is to introduce the fundamental ideas underlying Bayesian networks; to cover their uses in data analysis; and to demonstrate the use of related R packages using real data. The course will cover all three aspects of Bayesian networks: structure learning, parameter learning and inference. Advanced topics such as causal inference, handling missing data and model averaging will build on this material. Finally, some analyses from the literature will be replicated in R to provide examples on how to apply Bayesian networks to real-world data.

(Jan 23, 10-13; room G052)

  •  Relevant concepts graph theory: graphs, DAGs, cycles.
  •  Graphical separation and probabilistic independence.
  • Markov property and factorisation into local distributions.
  • The definition of Bayesian networks.
  • Markov blankets.
  • Parametric assumptions: discrete, Gaussian and conditional linear Gaussian Bayesian networks.

(Jan 23, 14.30-17.30; room G052)

  • Exact and approximate inference.
  • Junction trees.
  • Logic Sampling and Likelihood Weighting.
  • Diagnostic and prognostic models.
  • Naive Bayes and Tree-Augmented Naive Bayes classifiers.

(Jan 24, 10-13; room G052)

  • Causal inference
  • Missing data: Expectation Maximisation and Data Augmentation
  • Predictions, from a single model and from an ensemble.

(Jan 24, 14.30-17.30; room G052)

  • Structure learning.
  • Constraint-based, score-based, hybrid algorithms.
  • Common tests and scores.

(Jan 25, 10-13; room G052)

  • Graph priors (on the space of DAGs) in structure learning.
  • Parameter learning.
  • Model averaging.

(Jan 25, 14.30-17.30; room G052)

  • Protein signalling analysis from Sachs et al. (Sachs et al., Science, 2005).
  • Genomic association and prediction from Scutari et al. (Scutari et al., Genetics, 2014).
  • Modelling attitudes to business creation at universities (Garcia, Puga and Scutari, 2014, International Technology, Education and Development Conference)

Scarica le slide del corso qui