Aims

The EMUNE project pursues the following objectives:

1. To develop ML methods based on invertible neural networks for scalable parameter estimation and model selection based on large, but incomplete datasets.
2. To enable the analysis of host-pathogen interactions on the whole-body level from data collected in large-scale epidemiological studies.
3. To facilitate the understanding of immune-cell virus interactions on the tissue level from data collected using advanced imaging and single-cell sequencing technologies.
4. To validate the obtained methods and increasing our understanding of host-pathogen interactions by analyzing clinical and experimental data sets from viral infections.

Abstract

Host-pathogen interactions are complex biological processes that determine the outcome of infections. They are governed by the interplay of diverse factors across multiple scales in space and time, many of which are still poorly understood or even unknown. In particular for novel pathogens, the knowledge gaps are large and need to be closed quickly.

While imaging, serology and OMICS-technologies provide information in an unprecedented level of detail, and advanced organoid and tissue-culture systems allow for in-depth analysis and testing, the interpretation of the resulting data often remains challenging. Computational modelling using differential equations or hybrid discrete-continuum descriptions is here often the approach of choice. However, while these approaches allow for the integration of heterogeneous data and facilitate a mechanistic understanding of complex processes, the statistical inference is computationally demanding and does not scale to the steadily growing datasets from large clinical cohort studies and high-throughput technologies.

By using recently developed methods from machine learning (ML), including invertible neural networks (INNs) we aim to develop improvied methods for statistical inference in such highly complex systems. The application of these novel amortizing inference approaches to infectious disease research is limited by various theoretical and practical aspects, e.g. their current inability to handle missing data, to differentiate between competing hypotheses, or to flexibly integrate prior knowledge and constraints.

The overall aim of EMUNE is to build a framework for scalable statistical inference of host-pathogen interactions based on novel concepts from machine learning that account for the aforementioned challenges, and improves our understanding of infection and immune dynamics of specific pathogens.

Funding

EMUNE is funded by the German Ministry of Education and Research (BMBF) within the funding program “CompLS – Computational Life Sciences”.