Talks

Title: Bayesian inference of disease co-morbidity networks from electronic health records
Speaker: Guillermo Romero Moreno
Abstract:

Co-morbidity networks (a.k.a. Phenotypic disease networks) capture associations between morbidities that can be later used to explore disease progression, find typical co-morbidity patterns, etc. However, current methods for building co-morbidity networks suffer known biases and their statistical properties are not well defined or exploited. Here, we provide a new method for building comorbidity networks that correct some of the previous biases by disentangling the effects of co-morbidity and the ‘natural’ mechanisms for a morbidity to appear independently. This is achieved via a Bayesian approach that infers the latent target quantities from the data while also providing full statistical distributions of the inferred quantities and other network metrics. The Bayesian approach provides further benefits since i) it is based on a flexible and interpretable model that can be understood by domain experts; ii) it allows to inject domain knowledge in the models and priors; iii) it can generate alternative realisations of the original data, which can be particularly useful for comparing sub-populations at the level of raw co-occurrence of diseases. We apply this methodology to a primary care dataset and explore the benefits and uses of this methodology, while also comparing it to previous methods for building co-morbidity networks, such as relative risk or φ-correlations. We show that our methodology provides a more robust quantification of risk and that our results are more amenable to statistical comparison.

Title: Networks for smoking dynamics
Speaker: Adarsh Prabhakaran
Abstract:

Tobacco usage, one of the leading public health issues globally, has declined over the years due to multiple policies. Despite these policies, the decline of the number of smokers is slowing down, and tobacco use is still common. Over the years, multiple models have been developed to study the spread of smoking and to develop tobacco control strategies (mainly ODE - Ordinary Differential Equation models). However, these models do not consider all interactions between individuals observed empirically and their underlying network structure. In this talk, I will describe the network-based Agent-based model we developed to reproduce the social contagion process of tobacco use. Our results suggest that network structure is essential and that the observed dynamics from the ODE model are only similar to the network-based ABM only when the underlying network is fully connected, which is rarely the case. Further, by comparing multiple theoretical networks, we also show that networks with a similar average degree as the real-world network can be used to model smoking behaviour when complete information on the real-world network is unavailable. Additionally, our ABM can be used as a testbed to develop network-based intervention strategies and policies for tobacco control.

Title: Understanding the Clinical Pathways of Intensive Care Patients using AI
Speaker: Zonglin Ji
Abstract:

Clinical pathway analysis is pivotal in ensuring specialised, standardised, normalised and sophisticated therapy procedures. Understanding the patterns behind clinical pathways, especially what factors may cause them to deviate from expected standard procedures, are critical to providing decision support for clinicians and increasing the efficiency of medical care. Although existing pattern mining techniques can tell us about the sequence of medical behaviours in clinical pathways, few studies directly relate pathway patterns to the physiological data of patients. In this PhD project, we will explore how patient health data can affect and inform clinical pathways, especially at the early stage of hospital admission. We aim to predict the potential pathway that a new intensive care patient may take, focusing on possible deviations as they may indicate complications in their conditions and associated treatments. The ultimate goal of this project is to develop a novel framework that can dynamically predict patient pathways over time, thus providing continuous decision support for clinicians over ICU patients.

Title: Network models and machine learning for neuroimaging data
Speaker: Paola Galdi
Abstract:

Brain networks or connectomes are commonly used abstractions to model brain imaging data. Network nodes usually represent spatially contiguous regions of the brain, while edges model relationships between brain regions, that might reflect underlying anatomy (e.g., bundles of nerve fibres connecting neurons) or more complex interactions, such as correlated brain activity over time. In this talk, I will present four different brain network models derived from magnetic resonance imaging: functional connectivity networks, tractography-based connectomes, morphometric similarity networks and cortical meshes. I will then discuss a few examples of applications where features derived from brain networks are used to train predictive models and characterise clinical populations.

Title: Managing Sparsity in Formal Mathematics with User Tags and Network Models
Speaker: James Vaughan
Abstract:

Proof assistants are used for interactive development and verification of mathematical proofs with the assistance of automated tools. These proofs are highly modular and are frequently reused, but the quantity that are available makes it difficult to find those that are relevant during interactive proofs. In this presentation, I will discuss how we use network science to understand the structure of the dependency network that emerges between proofs in the Isabelle proof assistant. In particular, we use stochastic block models and side-information in the form of user tags to combat the sparsity of the network and improve the prediction of new dependencies. In so doing we aim to improve the level of automation in Isabelle, thereby enabling the user to tackle more complex proofs.

Title: The Usage and Improvement of Neurosymbolic AI for Biomedical Applications
Speaker: Lauren DeLong
Abstract:

Neurosymbolic Artificial Intelligence (AI) describes the combination of logic and rule-based approaches with deep learning. Often, the goal of neurosymbolic AI is to achieve comparable performance to current deep learning methods while simultaneously maintaining interpretability; this makes neurosymbolic AI both practically and ethically suitable for biomedical applications. However, few previous studies have attempted to apply neurosymbolic AI within the biomedical domain. Furthermore, many neurosymbolic approaches have unique abilities that neither symbolic nor neural approaches had alone. Many of these characteristics fit the unique challenges imposed by biomedical data. For example, some neurosymbolic approaches have the ability to represent meaningful relationships between data types, which could be used as a novel way to handle multimodal biomedical data fusion. Additionally, other neurosymbolic methods use a neural network to learn domain-specific rules; this opens the possibility to mine patterns from multi-relational biomedical data. The aims of this thesis, therefore, are to produce some of the first studies using neurosymbolic AI for biomedical applications as well as demonstrate ways to utilize the unique abilities of neurosymbolic methods for common challenges surrounding biomedical data.

Title: Formalising Haag-Kastler Nets In Higher-Order Logic
Speaker: Richard Schmoetten
Abstract:

Despite the enormous experimental success of quantum field theory (QFT) and its standard model of particle physics, its logical grounding remains ambiguous. Differing tools and methods are introduced more or less rigorously, and consequently the theory as a whole has no formal axiomatic basis. Algebraic QFT (AQFT) is an attempt to redress this deficiency, and my PhD is about enriching it with interactive and automated reasoning, in the guise of Isabelle/HOL.

Title: Resources and Process Compositions
Speaker: Filip Smola
Abstract:

In this talk I will highlight some of the progress I have made over the past year on my formal theory of resources and process compositions. This progress includes filling gaps in proofs, refining how resources are represented and introducing new concepts to the mechanised theory. With one of these concepts I seek to enrich resources with more information about any non-determinism they contain, while with another I seek to formally talk about the behaviour of processes.

Title: Improving the explainability of machine learning techniques in the healthcare domain
Speaker: Jorge Gaete Villegas
Abstract:

The critical nature of medical tasks makes explainability an essential quality for any support system in healthcare. Despite various techniques to provide explainable ML models, challenges still exist. Issues such as model selection, interpretation of results or user interaction with the models are important areas of research to achieve more understandable models and improve their adoption. In this talk, I will present our approach to tackle some of these issues, our current work and its application to patients in the Intensive Care Unit and plans for the rest of my PhD. Special attention will be paid to the use of community detection as an alternative to traditional clustering, and to the potential of leveraging our current results by using Probabilistic Logic Programming to create explainable predictive models.

Title: Understanding Processes
Speaker: Filip Smola
Abstract:

In this talk I will give a high-level overview of my work on mechanising linear resources and process models. I will focus on what we can express and do with these models rather than how the mechanisation is set up. This includes how we can use specific kinds of resources and basic actions to express different domains, as well as tools for formally relating such domains. The simpler applications I will demonstrate are built around automatically deriving information about complex compositions of processes in domains inspired by two simulation games. For more interesting applications I'll go over how process compositions relate to event sequences. And to conclude I'll highlight two pieces of work currently in progress: expressing distributions in non-deterministic resources and a labelled process transition system.