Short Talks Abstracts

Pariya Behrouzi 
Email: pariya.behrouzi@wur.nl
Affiliation: Wageningen University, NL
Construction of High-resolution Linkage Maps Using Discrete Graphical Models
Linkage maps are important for fundamental and applied genetic
research. In this talk, we introduce an algorithm to construct high-quality and
high-density linkage maps for diploid and polyploid species. We employ a sparse
Gaussian copula graphical model and the nonparanormal skeptic approach to
construct linkage maps. We compare our method with other available method
when the data are clean and contain no missing observations and when data are
noisy and incomplete. In addition, we implement the method on real genotype
data of barley and potato. We have implemented the method in the R package ”netgwas”
which is freely available at CRAN.

Denise De Gaetano 
Email: contact@denisedegaetano.com
Affiliation: MCAST
Extracting the Value of Big Data
Despite the large volumes of data, companies still struggle to access, manage and extract the information that their day-to-day processes generate. The growth of IT systems has provided these same companies, the ability to capture this potentially valuable data, within a number of applications, databases and organizations. Apart from a strong IT infrastructure, changes in the board and management, within a company also need to be carried out. This allows a number of departments to work in coordination and help ensure success of the utilization of the data at hand. Understanding what is required out of the data, is the first step to generate the best results from the company and customer data. The strategy is to understand how the information can enable an improvement in the business.

Arief Gusnanto
Email: a.gusnanto@leeds.ac.uk
Affiliation: Department of Statistics, School of Mathematics, University of Lees
Transcriptomic and Genomic Networks
Correlation network is an important tool in bioinformatics to find clusters of genes that are highly correlated in their expressions. The network can be used as a screening method to identify candidate biomarkers and therapeutic targets. This methodology has been successfully implemented in many biological contexts including cancer. While the interpretation may be natural in the context of gene expression data, network of copy number alterations (CNA) is not so straightforward because the data are in segments. CNA are structural variations in the human genome where some regions have more or less copy number than the normal two copies. Since the alterations happen in segments, the data exhibit stronger correlations than gene expression data and the correlations are in 'blocks'. The standard method is no longer adequate for an intuitive interpretation. This talk will describe the research problem, challenges, and our efforts so far in dealing with data from lung cancer patients. This is currently an ongoing joint work with (in alphabetical order) Mohammed Alshahrani, Luisa Cutillo, Charles Taylor, Peter Thwaites, and Henry Wood.

Rafiazka M. Hilman
Email: Hilman_Rafiazka@phd.ceu.edu
Affiliation: Central European University (CEU)
The Dynamics of Mass and Elite in Dutch Dividend Tax Discourse<
Political sphere in the Netherlands has been passing through torturous way due to the legislation process on Amendment to the Dividend Tax Act 1965 for the last 12 months. Among others, the government proposal on the abolishment of dividend tax (dividendbelasting) becomes the central point. Coalition parties who sponsor this bill, VVD, D66, CDA, and CU, deal with a lot of critiques from opposition in the parliament (Tweede Kamer).
There are three enthralling observations to make in the introduction of this bill. First, it creates ideological distance among central-right coalition parties in which coalition partners D66, CDA, and CU attempt to minimise the political sentiment caused by VVD’s main agenda on the abolishment of dividend tax. On this side, the improvement of investment climate serves as a shield. Secondly, it induces ideological proximity among opposition parties in the parliament where cross-spectrum stands on the same rejection platform by questioning policy cost at € 2 billion. Last, this political ambiguity leads to diverged public perception related to the importance of public spending over private incentive.
It is the central interest of this research to identify the alignment between public perception and elite discourse captured during the parliamentary debate session. Synthesis and analysis are made in response to two questions: How does social network reflect interactions between political elites, political parties, and mass in dividend tax discourse in the Netherlands? How does the political ambiguity evolve amidst the dynamics of dividend tax discourse?
In order to portray public perception towards elite interaction represented by political key players, social network data are extracted from Twitter in. On top of that minute of meetings recorded during parliament debate session are filtered out to construct the context and sentiment fragmentation among elites. Data are collected using Twitter API service during 4 weeks-period in October 2018. This period is selected to enable the tail-end of dividend tax issue as the government decided to withdraw the plan in the beginning of October 2018. Meanwhile, parliament record is analysed from the initial discussion in November 2017 to October 2018.
The research proceeds as follows: first, the methodology comprising data and model are presented. Next, ideological spectrum and coalition formation becomes a foundation of the following discussion on political dynamics surrounding dividend tax discourse. The final part of the article concludes and discusses the results in terms of network structure and network properties.

Martin Lopez Garcia
Email: m.lopezgarcia@leeds.ac.uk
Affiliation: Department of Applied Mathematics, School of Mathematics, University of Leeds, UK
On the exact analysis of stochastic epidemic processes on networks
I will show in this talk how to analyse the SIR epidemic model in an exact way when the population under study is formed by a small highly heterogeneous group of N individuals, represented by means of a network. This approach, which amounts to the analysis of the exact 3^N-states continuous-time Markov chain (CTMC), makes special focus on algorithmic aspects, and requires a creative organization of the space of states {S,I,R}^N of the CTMC. The analysis of the epidemic dynamics is carried out in terms of a number of summary statistics for the disease: (i) the length and size of the outbreak; (ii) the maximum number of individuals simultaneously infected during the outbreak; (iii) the fate of a particular individual within the
population; and (iv) the number of secondary cases caused by a certain individual until she/he recovers. I will illustrate this methodology by studying the spread of the nosocomial pathogen Methicillin-resistant Staphylococcus Aureus among the patients within an intensive care unit (ICU). The interest here is in analysing the effectiveness of different control strategies which intrinsically incorporate heterogeneities among the patients within the ICU.
References:
M. López-García (2016) Stochastic descriptors in an SIR epidemic model for heterogeneous individuals in small networks. Mathematical Biosciences 271: 42-61.
A. Economou, A. Gómez-Corral, M. López-García (2015) A stochastic SIS epidemic model with heterogeneous contacts. Physica A: Statistical Mechanics and its Applications 421: 78-97.

Marija Mihova 
Email: marija.mihova@finki.ukim.mk
Affiliation: Ss. Cyril and Methodius University, FINKI, Skopje
Efficient Algorithm for Finding all Minimal Path Vectors in Two-terminal Flow Network
Minimal path and minimal cut vectors are usually used for computing the reliability of a two- terminal flow network with discrete set of possible capacities of its arcs. This work unites the max-flow theory of two-terminal vectors and the theory of minimal path vectors in multi-state systems. Based on obtained theoretical results, we have designed an algorithm for computing all minimal path vectors for a given level d.

Rebeka O. Szabo 
Email: o.sz.rebu@gmail.com
Affiliation: CEU
The micro-dynamic nature of team interactions.
Teams have become a popular organization form since well-functioning task-focused groups
are basic pillars of successful organizations. While there is much interest in contemporary social
science in understanding team processes that lead to efficiency, most of these researches rely
heavily on self-reported data yielding static and potentially biased information and tends to
overlook actual interaction processes. We propose a novel approach that allows portraying a
nuanced, complex picture of problem-solving group behaviour by measuring performance
dynamics as it evolves in real-time, in a controlled environment. The research aims to explore
how collaboration networks of small project teams evolve across time and team members, and
how it relates to successful task performance. We investigate interaction patterns in escape
rooms, where all teams are video recorded during the task-solving process in the same
experimental environment. We expected and confirmed that homogeneous distribution of
interaction ties across time and team members fosters successful problem-solving. Concerning
the impact of the initial social roles on the dynamics of the interaction pattern, we hypothesized
that flexible, less hierarchical team structures favour for problem-solving. In the case of the
teams with random composition, the development of a new social structure during the dynamic
performance of an unstructured task is expected to entail more tensions with the conversation
rules than otherwise. This research aims to advance the new science of teams' by focusing on
the network micro-mechanisms that allows us to treat teams as dynamic, adaptive, task-
performing systems.

Valeria Policastro 
Email: valeria.policastro@gmail.com
Affiliation: IAC Istituto per le Applicazioni del Calcolo "Mauro Picone", CNR Naples, Italy
ROBIN: an R package for validation of community robustness 
In network analysis, many community detection algorithms have been developed. However, their applications leave unaddressed one important question: the statistical validation of the results. We present ROBIN (Robustness In Network), an R package that gives a statistical answer to the validation of the community structure by looking at the robustness of the network. The package implements a methodology presented in a previous paper that detects if the community structure found by a detection algorithm is statistically significant or is a result of chance, merely due to edge positions in the network. The software performs a perturbation strategy and runs a null model to build a set of procedures based on the Variation of Information as a clustering distance. In particular, it provides a procedure to examine the stability of the partition recovered against random perturbations of the original graph structure, a routine to compare different detection algorithms applied to the same network and a graphical interactive representation of networks. The package is useful not only to determine whether the obtained clustering departs significantly from the null model, but also to discover which algorithm better fits for the network of interest.

Posters Abstracts

Helena Andres-Terre 
Email: helenaandres@gmail.com
Affiliation: University of Cambridge
The introduction of single cell RNA-seq data was a major breakthrough in the field of biology, and particularly useful in research areas such as comparative transcriptomics or disease studies. It allows the characterization of gene expression levels for individual cells, and the potential to observe different stages of the Stem cell differentiation process.
These datasets are known to be sparse and highly dimensional, with a large number of genes describing each cell. In order to interpret the experimental results, one of the main objectives is to identify the most relevant features of the underlying processes. After a first cut on the number of accounted genes based on their variability, current techniques use dimensionality reduction methods such as Principal Component Analysis (linear) or tSNE (non-linear). The new components are then used for plotting, performing further analysis on classification tasks or to describe differentiation processes. While these methods have been proved valid to address the aforementioned challenges, they also present some restrictions when trying to characterize the middle states of differentiation.
We present an unsupervised Machine Learning technique for dimensionality reduction of single cell data. We used Variational Auto-Encoders to extract a number of significant components that characterize individual cells based on their gene expression, using a deep learning bottle-neck approach.
The variational nature of this technique allows for certain levels of stochasticity in the original data, while learning an encoded representation that can be used to reconstruct and generate new samples.
The methods that are based on linear dimensionality reduction techniques often require strong assumptions and constraints; for instance locally homogeneous distribution of samples in the high dimensional space, or not accounting for low variance dimensions. Variational Auto-Encoders (VAEs) provide an en- coding capable to identify and separate among relevant features in the dataset.
VAEs can also be used to identify sets of relevant genes or drivers of the different processes captured by the latent encoding. Their generative potential also allows the exploration of the latent space, which can lead to defining a theoretical energy landscape that describes trajectories of differentiation.

Annamaria Carissimo 
Email: a.carissimo@na.iac.cnr.it
Affiliation: Istituto per le Applicazioni del Calcolo "M.Picone", Naples, CNR
SusNet: a global retinal co-expression network
Network analysis provides a useful framework to visualize and analyze complex biological problems. In biological networks, transcripts, genes or proteins are represented as nodes, and relationships between them as edges. These interactions can be reconstructed by inference methods starting from expression profiles. Co-expression networks use the transcriptional concordance of two gene expression profiles to build undirected graph representations of the biological system under observation. In our study, we generated a co-expression network of the adult porcine retina, SusNet,  by calculating pairwise gene correlation among 47 Sus Scrofa Large White retina samples sequenced by RNA-Seq.  We showed that SusNet captures the pan-retinal regulatory structure associated to retina-specific TFs. We further showed that mapping differentially expressed genes (DEGs) following somatic repression of the rod-specific gene Rhodopsin on SusNet enables the identification of a sub-GRN operating in a subcellular compartment.

Antonella Iuliano
Email: a.iuliano@tigem.it
Affiliation: Telethon Institute of Genetics and Medicine (TIGEM), Pozzuoli, Italy
COSMONET
We present an R package called COSMONET where screening-network methods are implemented for the prediction of survival outcome in cancer patients. The novelty is the combination of different types of screenings (biomedical-driven, data-driven, the union of both) and network-regularized Cox methods. This approach allows to improve the prediction capabilities, to discriminate patients in high-and low-risk groups using few potential biomarkers, and to help clinicians in the management of patients.

Gamze OZEL KADILAR 
Email: gamzeozl@hacettepe.edu.tr
Affiliation: Hacettepe University, Ankara, Turkey
Graphical Markov Models with an Application on Traffic Accident Data
Undirected graphical models are widely used for modeling, visualization, inference, and exploratory analysis of multivariate data with wide-ranging applications. Graphical models are models based on graphs in which nodes represent random variables, and the edges represent conditional independence assumptions. Hence they provide a compact representation of joint probability distributions. Graphical Markov models started to be developed after 1970 as special subclasses of log-linear models for contingency tables and of joint Gaussian distributions, where conditional independence constraints are imposed such that conditioning is on all the other variables. The study of these models is an active research area, with many questions still open. In this study, graphical Markov models are described. Then, interpretations are illustrated with an application based on traffic accident data of Turkey. Furthermore, some of the more recent, important results for sequences of regressions are summarized.
Keywords: graphical models; Markov property; categorical data analysis, traffic accident data.

Monika Krzak
Email: m.Krzak@na.iac.cnr.it
Affiliation: Istituto per le Applicazioni del Calcolo "M.Picone", Naples, CNR
Identification of cell subpopulations using ensMAP-DP approach
Single-cell RNA sequencing (scRNAseq) has emerged as an important technology that allows profiling gene expression at single-cell resolution. The great potential of this technique lies in the possibility to infer cellular diversity within the same organ, tissue or group of cells of interest. In the last years, several studies have been carried out for identifying novel or known cell populations. Despite the great collection of available methods, an accurate detection of cell subpopulations remains unresolved and several issues are still open. For example, most of the algorithms require to provide a fixed number of
desired subpopulations. This might be a drawback when no prior knowledge about cell population is available or when the aim is to identify novel subtypes of cells. Motivated by these reasons, we developed a new method, ensMAP-DP, that uses probabilistic mixture modeling to reveal latent cell subpopulations. The method consists of several steps: first, it selects a number of most relevant features (i.e. so-called highly variable genes), then applies tSNE and performs MAP-DP clustering on a given number of
components, finally the clustering solutions corresponding to different tSNE projections are combined into a consensus clustering using a meta-clustering algorithm. In this work, we demonstrate the superior performance of our approach to other widely used methods designed to infer the cellular heterogeneity in scRNAseq data.

Margherita Mutarelli 
Email: mutarelli@tigem.it
Affiliation: TIGEM
A Graph Database Resource for Disease Gene Annotation and Prioritization
The identification of the molecular cause of genetic diseases is one of the fundamental questions of medical research. Despite the availability of high-throughput techniques like Whole Exome and Whole Genome Sequencing that made now  feasible to genotype patients at an unprecedented level of depth, still a high proportion of cases remain unsolved. We are building a gene and disease network based on the annotated phenotypes present in patients to improve candidate causative gene prioritization.

Alice Tapper 
Email: mmata@leeds.ac.uk
Affiliation: University of Leeds
Inference on multilayer networks with latent layers
Online social networks often represent one layer in a larger multilayer network of connections between people. Motivated by this, SIS dynamics occurring on two-layer systems with one visible layer and one latent layer are studied. A range of network structures and epidemic parameters are explored using mean field approximations and simulation, with an aim to identify cases where inference is inaccurate if solely the visible layer is analysed.