Using Epigenetically-Inspired Connectionist Models to Provide Transparency In The Modelling of Human Visceral Leismaniasis

Project Description

A connectionist model (CM) is one where computation and intelligence emerge from the interactions of individually simple non-linear elements. CMs have been at the forefront of computational intelligence for decades and are becoming an ever more pervasive computational tool. However, with the increased abundance of CMs, and their application in areas of �high risk� such as medical diagnostics and biological modelling, it is increasingly important to understand why certain decisions have been made in order to provide transparency and confidence in their decision-making processes. At present, most CMs provide no transparency of decision making process and are considered to be �black box� models.
Imagine instead, a CM which is capable of learning complex behaviours and simultaneously providing an explanation of exactly why certain decisions have been made without prohibitively high computational overheads; a CM which can provide such information during the optimisation process to ensure that the rules and behaviours it is developing are in line with the domain knowledge; a CM which can provide novel insight and knowledge discovery within an application domain, in turn providing evidence to support its decision-making process and giving confidence in its behaviour. Epigenetically-inspired CMs (EICMs) have the capacity to engender such properties

Human visceral leismaniasis (HVL) is a disease caused by protozoan parasites of the Leishmania genus, and is the second largest parasitic killer after malaria with 50000 � 90000 new cases annually [L17]. Infection with HVL is characterised by a complex immune response involving antigens, T-cells and B-cells. However, the interactions between these entities during HVL infection are not well understood, and this limits the potential for developing treatments.
With the overall goal to produce knowledge pertinent to the immune response in relation to HVL, this proposed research has two key foci. First, it will focus on the functionality of EICMs, building upon previous work [T13a,T13b,T13c,L14,T15,T16a,T17a,T17b] to adapt them to autonomously provide explanations of their own decision making, allowing on the fly transparency of execution. The second focus is to apply this adapted EICM and the techniques developed in this work to provide key scientific insight into the immune response in HVL. Such a task is particularly well suited to epigenetically-inspired connectionist models, as the biological immune response is a product of many interacting biological pathways over varying time scales and underpinned by various biological epigenetic mechanisms. Existing modelling approaches attempt to capture biological observations within a computational model without focusing on how the model is interpreting the data, which in turn reduces the confidence that can be obtained from the model. The EICMs will be used to emulate existing biological models, by modellling the biological data underpinning them, to provide key scientific insight into HVL development which in turn can be used as evidence to update the biological understanding of such processes. The data and existing models of the immune response to HVL will be provided by Simomics, a world leading company in modelling disease development who focus on transparent modelling in biological simulations.