In this paper, we provide explicit upper bounds on some distances between the (law of the) output of a random Gaussian neural network and (the law of) a random Gaussian vector. Our main results concern deep random Gaussian neural networks with a rather general activation function. The upper bounds show how the widths of the layers, the activation function, and other architecture parameters affect the Gaussian approximation of the output. Our techniques, relying on Stein's method and integration by parts formulas for the Gaussian law, yield estimates on distances that are indeed integral probability metrics and include the convex distance. This latter metric is defined by testing against indicator functions of measurable convex sets and so allows for accurate estimates of the probability that the output is localized in some region of the space, which is an aspect of a significant interest both from a practitioner's and a theorist's perspective. We illustrated our results by some numerical examples.

Introduction: Connections among neurons form one of the most amazing and effective network in nature. At higher level, also the functional structures of the brain is organized as a network. It is therefore natural to use modern techniques of network analysis to describe the structures of networks in the brain. Many studies have been conducted in this area, showing that the structure of the neuronal network is complex, with a small-world topology, modularity and the presence of hubs. Other studies have been conducted to investigate the dynamical processes occurring in brain networks, analyzing local and large-scale network dynamics. Recently, network diffusion dynamics have been proposed as a model for the progression of brain degenerative diseases and for traumatic brain injuries. Methods: In this paper, the dynamics of network diffusion is re-examined and reaction-diffusion models on networks is introduced in order to better describe the degenerative dynamics in the brain. Results: Numerical simulations of the dynamics of injuries in the brain connectome are presented. Different choices of reaction term and initial condition provide very different phenomenologies, showing how network propagation models are highly flexible. Discussion: The uniqueness of this research lies in the fact that it is the first time that reaction-diffusion dynamics have been applied to the connectome to model the evolution of neurodegenerative diseases or traumatic brain injury. In addition, the generality of these models allows the introduction of non-constant diffusion and different reaction terms with non-constant parameters, allowing a more precise definition of the pathology to be studied.

In this talk we provide explicit upper bounds on some distances between the (law of the) output of a random Gaussian neural network and (the law of) a random Gaussian vector. Our main results concern deep random Gaussian neural networks, with a rather general activation function. The upper bounds show how the widths of the layers, the activation function and other architecture parameters affect the Gaussian approximation of the output. Our techniques, relying on Stein's method and integration by parts formulas for the Gaussian law, yield estimates on distances which are indeed integral probability metrics, and include the convex distance. This latter metric is defined by testing against indicator functions of measurable convex sets, and so allows for accurate estimates of the probability that the output is localized in some region of the space. Such estimates have a significant interest both from a practitioner's and a theorist's perspective.

This contribution outlines current research aimed at developing models for personalized type 2 diabetes mellitus (T2D) prevention in the framework of the European project PRAESIIDIUM (Physics Informed Machine Learn-ing-Based Prediction and Reversion of Impaired Fasting Glucose Management) aimed at building a digital twin for preventing T2D in patients at risk. Specifically, the modelling approaches include both a multiscale, hybrid computational model of the human metaflammatory (metabolic and inflammatory) status, and data-driven models of the risk of developing T2D able to generate personalized recommendations for mitigating the individ-ual risk. The prediction algorithm will draw on a rich set of information for training, derived from prior clinical data, the individual's family history, and prospective clinical trials including clinical variables, wearable sensors, and a tracking mobile app (for diet, physical activity, and lifestyle). The models developed within the project will be the basis for building a platform for healthcare professionals and patients to estimate and monitor the indi-vidual risk of T2D in real time, thus potentially supporting personalized prevention and patient engagement.

Motivation: Gene-disease associations are fundamental for understanding disease etiology and developing effective interventions and treatments. Identifying genes not yet associated with a disease due to a lack of studies is a challenging task in which prioritization based on prior knowledge is an important element. The computational search for new candidate disease genes may be eased by positive-unlabeled learning, the machine learning setting in which only a subset of instances are labeled as positive while the rest of the data set is unlabeled. In this work, we propose a set of effective network-based features to be used in a novel Markov diffusion-based multi-class labeling strategy for putative disease gene discovery. Results: The performances of the new labeling algorithm and the effectiveness of the proposed features have been tested on ten different disease data sets using three machine learning algorithms. The new features have been compared against classical topological and functional/ontological features and a set of network- and biological-derived features already used in gene discovery tasks. The predictive power of the integrated methodology in searching for new disease genes has been found to be competitive against state-of-the-art algorithms.Availability and implementation: The source code of NIAPU can be accessed at https://github. com/AndMastro/NIAPU. The source data used in this study are available online on the respective websites.

Introduction: Mathematical modeling has emerged as a crucial component in understanding the epidemiology of infectious diseases. In fact, contemporary surveillance efforts for epidemic or endemic infections heavily rely on mathematical and computational methods. This study presents a novel agent-based multi-level model that depicts the transmission dynamics of gonorrhea, a sexually transmitted infection (STI) caused by the bacterium Neisseria gonorrhoeae. This infection poses a significant public health challenge as it is endemic in numerous countries, and each year sees millions of new cases, including a concerning number of drug-resistant cases commonly referred to as gonorrhea superbugs or super gonorrhea. These drug-resistant strains exhibit a high level of resistance to recommended antibiotic treatments.MethodsThe proposed model incorporates a multi-layer network of agents' interaction representing the dynamics of sexual partnerships. It also encompasses a transmission model, which quantifies the probability of infection during sexual intercourse, and a within-host model, which captures the immune activation following gonorrhea infection in an individual. It is a combination of agent-based modeling, which effectively captures interactions among various risk groups, and probabilistic modeling, which enables a theoretical exploration of sexual network characteristics and contagion dynamics.ResultsNumerical simulations of the dynamics of gonorrhea infection using the complete agent-based model are carried out. In particular, some examples of possible epidemic evolution are presented together with an application to a real case study. The goal was to construct a virtual population that closely resembles the target population of interest.DiscussionThe uniqueness of this research lies in its objective to accurately depict the influence of distinct sexual risk groups and their interaction on the prevalence of gonorrhea. The proposed model, having interpretable and measurable parameters from epidemiological data, facilitates a more comprehensive understanding of the disease evolution.

The importance of faster drug development has never been more evident than in present time when the whole world is struggling to cope up with the COVID-19 pandemic. At times when timely development of effective drugs and treatment plans could potentially save millions of lives, drug repurposing is one area of medicine that has garnered much of research interest. Apart from experimental drug repurposing studies that happen within wet labs, lot many new quantitative methods have been proposed in the literature. In this paper, one such quantitative methods for drug repurposing is implemented and evaluated. DruSiLa (DRUg in-SIlico LAboratory) is an in-silico drug re- purposing method that leverages disease similarity measures to quantitatively rank existing drugs for their potential therapeutic efficacy against novel diseases. The proposed method makes use of available, manually curated, and open datasets on diseases, their genetic origins, and disease-related patho-phenotypes. DruSiLa evaluates pairwise disease similarity scores of any given target disease to each known disease in our dataset. Such similarity scores are then propagated through disease-drug associations, and aggregated at drug nodes to rank them for their predicted effectiveness against the target disease.

Most financial signals show time dependency that, combined with noisy and extreme events, poses serious problems in the parameter estimations of statistical models. Moreover, when addressing asset pricing, portfolio selection, and investment strategies, accurate estimates of the relationship among assets are as necessary as are delicate in a time-dependent context. In this regard, fundamental tools that increasingly attract research interests are precision matrix and graphical models, which are able to obtain insights into the joint evolution of financial quantities. In this paper, we present a robust divergence estimator for a time-varying precision matrix that can manage both the extreme events and time-dependency that affect financial time series. Furthermore, we provide an algorithm to handle parameter estimations that uses the "maximization-minimization" approach. We apply the methodology to synthetic data to test its performances. Then, we consider the cryptocurrency market as a real data application, given its remarkable suitability for the proposed method because of its volatile and unregulated nature.

Background: The immune response to adenoviral COVID-19 vaccines is affected by the interval between doses. The optimal interval is unknown.Aim: We aim to explore in-silico the effect of the interval between vaccine administrations on immunogenicity and to analyze the contribution of pre-existing levels of antibodies, plasma cells, and memory B and T lymphocytes.Methods: We used a stochastic agent-based immune simulation platform to simulate two-dose and three-dose vaccination protocols with an adenoviral vaccine. We identified the model's parameters fitting anti-Spike antibody levels from individuals immunized with the COVID-19 vaccine AstraZeneca (ChAdOx1-S, Vaxzevria). We used several statistical methods, such as principal component analysis and binary classification, to analyze the correlation between pre-existing levels of antibodies, plasma cells, and memory B and T cells to the magnitude of the antibody response following a booster dose.Results and conclusions: We find that the magnitude of the antibody response to a booster depends on the number of pre-existing memory B cells, which, in turn, is highly correlated to the number of T helper cells and plasma cells, and the antibody titers. Pre-existing memory T cytotoxic cells and antibodies directly influence antigen availability hence limiting the magnitude of the immune response. The optimal immunogenicity of the third dose is achieved over a large time window, spanning from 6 to 16 months after the second dose. Interestingly, after any vaccine dose, individuals can be classified into two groups, sustainers and decayers, that differ in the kinetics of decline of their antibody titers due to differences in long-lived plasma cells. This suggests that the decayers may benefit from a tailored boosting schedule with a shorter interval to avoid the temporary loss of serological immunity.

Primary biliary cholangitis (PBC) is a chronic, cholestatic, immune-mediated, and progressive liver disorder. Treatment to preventing the disease from advancing into later and irreversible stages is still an unmet clinical need. Accordingly, we set up a drug repurposing framework to find potential therapeutic agents targeting relevant pathways derived from an expanded pool of genes involved in different stages of PBC. Starting with updated human protein–protein interaction data and genes specifically involved in the early and late stages of PBC, a network medicine approach was used to provide a PBC “proximity” or “involvement” gene ranking using network diffusion algorithms and machine learning models. The top genes in the proximity ranking, when combined with the original PBC-related genes, resulted in a final dataset of the genes most involved in PBC disease. Finally, a drug repurposing strategy was implemented by mining and utilizing dedicated drug–gene interaction and druggable genome information knowledge bases (e.g., the DrugBank repository). We identified several potential drug candidates interacting with PBC pathways after performing an over-representation analysis on our initial 1121-seed gene list and the resulting disease-associated (algorithm-obtained) genes. The mechanism and potential therapeutic applications of such drugs were then thoroughly discussed, with a particular emphasis on different stages of PBC disease. We found that interleukin/EGFR/TNF-alpha inhibitors, branched-chain amino acids, geldanamycin, tauroursodeoxycholic acid, genistein, antioestrogens, curcumin, antineovascularisation agents, enzyme/protease inhibitors, and antirheumatic agents are promising drugs targeting distinct stages of PBC. We developed robust and transparent selection mechanisms for prioritizing already approved medicinal products or investigational products for repurposing based on recognized unmet medical needs in PBC, as well as solid preliminary data to achieve this goal.

The method of simulated quantiles is extended to a general multivariate framework and to provide sparse estimation of the scaling matrix. The method is based on the minimisation of a distance between appropriate statistics evaluated on the true and synthetic data simulated from the postulated model. Those statistics are functions of the quantiles providing an effective way to deal with distributions that do not admit moments of any order like the α–Stable or the Tukey lambda distribution. The lack of a natural ordering represents the major challenge for the extension of the method to the multivariate framework, which is addressed by considering the notion of projectional quantile. The SCAD 1–penalty is then introduced in order to achieve sparse estimation of the scaling matrix which is mostly responsible for the curse of dimensionality. The asymptotic properties of the proposed estimator have been discussed and the method is illustrated and tested on several synthetic datasets simulated from the Elliptical Stable distribution for which alternative methods are recognised to perform poorly.

Background: The aim of the present paper is to construct an emulator of a complex biological system simulator using a machine learning approach. More specifically, the simulator is a patient-specific model that integrates metabolic, nutritional, and lifestyle data to predict the metabolic and inflammatory processes underlying the development of type-2 diabetes in absence of familiarity. Given the very high incidence of type-2 diabetes, the implementation of this predictive model on mobile devices could provide a useful instrument to assess the risk of the disease for aware individuals. The high computational cost of the developed model, being a mixture of agent-based and ordinary differential equations and providing a dynamic multivariate output, makes the simulator executable only on powerful workstations but not on mobile devices. Hence the need to implement an emulator with a reduced computational cost that can be executed on mobile devices to provide real-time self-monitoring. Results: Similarly to our previous work, we propose an emulator based on a machine learning algorithm but here we consider a different approach which turn out to have better performances, indeed in terms of root mean square error we have an improvement of two order magnitude. We tested the proposed emulator on samples containing different number of simulated trajectories, and it turned out that the fitted trajectories are able to predict with high accuracy the entire dynamics of the simulator output variables. We apply the emulator to control the level of inflammation while leveraging on the nutritional input. Conclusion: The proposed emulator can be implemented and executed on mobile health devices to perform quick-and-easy self-monitoring assessments.

Background: The aim of a recent research project was the investigation of the mechanisms involved in the onset of type 2 diabetes in the absence of familiarity. This has led to the development of a computational model that recapitulates the aetiology of the disease and simulates the immunological and metabolic alterations linked to type-2 diabetes subjected to clinical, physiological, and behavioural features of prototypical human individuals. Results: We analysed the time course of 46,170 virtual subjects, experiencing different lifestyle conditions. We then set up a statistical model able to recapitulate the simulated outcomes. Conclusions: The resulting machine learning model adequately predicts the synthetic dataset and can, therefore, be used as a computationally-cheaper version of the detailed mathematical model, ready to be implemented on mobile devices to allow self-assessment by informed and aware individuals. The computational model used to generate the dataset of this work is available as a web-service at the following address: http://kraken.iac.rm.cnr.it/T2DM.

The paper proposes an approach to deal with the day by day dynamic behaviour of Oil & Gas assets, providing support for optimized decisions on wells and facilities. The approach is based on: o A set of software agents, trained with a machine learning approach to understand the health status of the components of the reservoir/well/plant system and capable of proposing optimization actions for the corresponding subsystem; o An inter-agent negotiation approach, capable of evaluating the optimization actions of the single agents in the wider picture of the overall optimization of the producing asset. The paper will describe how this approach has been implemented, as well as an example application.

The ongoing COVID-19 pandemic still requires fast and effective efforts from all fronts, including epidemiology, clinical practice, molecular medicine, and pharmacology. A comprehensive molecular framework of the disease is needed to better understand its pathological mechanisms, and to design successful treatments able to slow down and stop the impressive pace of the outbreak and harsh clinical symptomatology, possibly via the use of readily available, off-the-shelf drugs. This work engages in providing a wider picture of the human molecular landscape of the SARS-CoV-2 infection via a network medicine approach as the ground for a drug repurposing strategy. Grounding on prior knowledge such as experimentally validated host proteins known to be viral interactors, tissue-specific gene expression data, and using network analysis techniques such as network propagation and connectivity significance, the host molecular reaction network to the viral invasion is explored and exploited to infer and prioritize candidate target genes, and finally to propose drugs to be repurposed for the treatment of COVID-19. Ranks of potential target genes have been obtained for coherent groups of tissues/organs, potential and distinct sites of interaction between the virus and the organism. The normalization and the aggregation of the different scores allowed to define a preliminary, restricted list of genes candidates as pharmacological targets for drug repurposing, with the aim of contrasting different phases of the virus infection and viral replication cycle.

In this study, the results of 46170 simulations corresponding to the same number of virtual subjects, experiencing different lifestyle conditions, are analysed for the construction of a statistical model able to recapitulate the simulated dynamics. Investigation about the mechanisms involved in the onset of type 2 diabetes in absence of familiarity is the focus of a research project which has led to the development of a computational model that recapitulates the aetiology of the disease. The model simulates the metabolic and immunological alterations related to type-2 diabetes associated to several clinical, physiological and behavioural characteristics of representative virtual patients.

Investigation about the mechanisms involved in the onset of type 2 diabetes in absence of familiarity is the focus of a research project which has led to the development of a computational model that recapitulates the aetiology of the disease. The model simulates the metabolic and immunological alterations related to type-2 diabetes associated to several clinical, physiological and behavioural characteristics of representative virtual patients. In this study, the results of 46170 simulations corresponding to the same number of virtual subjects, experiencing different lifestyle conditions, are analysed for the construction of a statis- tical model able to recapitulate the simulated dynamics. The resulting machine learning model adequately predicts the synthetic data and can therefore be used as a computationally- cheaper version of the detailed mathematical model, ready to be implemented on mobile devices to allow self assessment by informed and aware individuals.

GOLIAT is an offshore production field that spans from the subsea wells up to a complete process plant installed on a FPSO. Due to the comprehensive instrumentation installed on the plant, it is the perfect case study to test an innovative agent based software architecture able to support production management. The modularity and the scalability provided by the agent based architecture guarantees the applicability of the method to any part of the plant. Each agent is in charge of supervising a specific or a group of equipment and is fed by the real-time data coming from the field. These data are then analysed through Machine Learning and Deep Learning algorithms which are incorporated within the agents. The machine learning algorithms estimate the current state of the equipment and provide a set of KPIs in order to understand both the production efficiency and the health status of the machines. Furthermore, learning from the observations of the state transition paths which happened in the past, the agents are capable of predicting the most likely future state. The latter capability is fundamental to prevent unplanned shutdowns and optimize the maintenance plans. On the basis of the estimated current state, each agent can also provide a list of actions targeted to maximize the efficiency from an "equipment" point of view. The actions coming from all the agents can then be collected and negotiated in order to maximize the production from a "plant" point of view. The negotiating algorithms are implemented in a super-agent that can support a human operator in the day-by-day management tasks of the plant. Even though the negotiating capabilities will be implemented in the future version of the application, the modular nature of the system guarantees an easy integration of the super-agent inside the agent's framework. The paper will present the results of the agent framework in terms of the robustness of state estimation and the correctness of the computed KPIs.

This paper introduces a dominance test that allows to determine whether or not a financial institution can be classified as being more systemically important than another in a multivariate framework. The dominance test relies on a new risk measure, the NetCoVaR that is specifically tailored to capture the joint extreme co-movements between institutions belonging to a network. The asymptotic theory for the statistical test is provided under mild regularity conditions concerning the joint distribution of asset returns which is assumed to be elliptically contoured. The proposed risk measure and risk measurement framework is used to analyse the US financial system during the recent Global Financial Crises. In the empirical analysis, the returns are assumed to be Elliptically Stable distributed and the estimation is carried out through the Sparse Multivariate Method of Simulated Quantiles, handling both the lack of an analytic expression for the probability density function and the potential high-dimensionality of the problem.