MuMoSim: Machine Learning Supported Multi-Model Simulator for Infection Research

www.leibniz-hki.de References κN MuMoSim: Machine Learning Supported Multi-Model Simulator for Infection Research Anastasia Solomatina1, Kerstin Hünniger2,3, Yann Bachelot1,4, Jana Wilms1,4, Oliver Kurzai2,3, Marc Thilo Figge1,5 1 Applied Systems Biology, Leibniz Institute for Natural Product Research and Infection Biology, Hans-Knöll-Institute (HKI), Jena, Germany 2 Fungal Septomics, Septomics Research Center, Leibniz Institute for Natural Product Research and Infection Biology, Hans-Knöll-Institute (HKI), Jena, Germany 3 Institute for Hygiene and Microbiology, University of Würzburg, Würzburg, Germany 4 Faculty of Biological Sciences, Friedrich Schiller University, Jena, Germany 5 Institute of Microbiology, Faculty of Biological Sciences, Friedrich-Schiller-University, Jena, Germany [1] Hünniger et al. 2014. PLOS Comput Biol. 10(2) [2] Lehnert et al. 2021. Sci Rep. 11(1) [3] Lehnert et al. 2015. Front Microbiol. 6:608 [4] Haykin 1994. Prentice Hall PTR [5] Akiba et al. 2019. In KDD [email protected] Abstract anti-coagulated whole blood pathogen C.albicans Monocytes Neutrophils Phagocytosis Phagocytosis Phagocytosis Intracellular killing Intracellular killing Extracellular killing Immune escape Statebased model Conclusions and outlook Exp. data association assay 0.0076 0.0039 0.0019 0.4917 0.4584 0.2283 0.3482 association & survival assay 0.0036 0.0110 0.0010 0.5801 0.2081 0.2368 0.2272 association & survival & killing assay 0.0052 0.0026 0.0006 0.5186 0.0625 0.2220 0.3292 ϕN ϕM ρ γ κN κM κEK Uncertainty: small medium large •The gradual increase of model complexity from SBM to hABM enables us to describe chemokineinduced guidance of cells to sites of infection. •The parameter inference for the SBM module allows prediction for experimental design. •Surrogate modeling enables parameter inference for computationally expensive ABM simulation. • MuMoSim is a general framework that aims to quantify host-pathogen interactions for various conditions (bacterial/fungal infections in patients cohorts) and to predict treatment strategies. Systemic fungal infections •Candida albicans is a major cause of nosocomial systemic infections • Patients with neutropenia and under immunosuppressive therapy are at high risk for an infection • Two main routes to enter the bloodstream: GI tract Medical devices Human whole-blood infection assay[1] Clinical blood samples [2] Blood from healthy volunteers analyze the interplay of fungi with host immunity during the initial infection phase[1] Ex vivo human whole-blood assay: Agent-based model • Stochastic temporal model • Well-mixed environment association & survival assay ϕN Interactions: κEK,γ ϕN ϕM ρ ρ • Posterior distributions vs. the ground truth • Wasserstein distance used as a metric •Killing assay: % pathogens killed by neutrophils & monocytes • Stochastic spatiotemporal model •3D environment 1 of blood •Contact-dependent interactions → μL Experimental data: Experimental design: MonocytesNeutrophils Pathogens alive killed immune-evasive ϕM κM t0 t1 t2 t3 t4 t5 α0 α1 α2 α3 Migration: random walk Posterior distributions for kinetic parameters & best-fit values[3]: 1 mm Input layer Hidden layers Output layer … … decrease of uncertainty in the killing rate of neutrophils κN Killing assay data for SBM: cytokines concentration Experimental data: • ML surrogate model for parameter inference •Multilayer perceptron[4] as a surrogate model •Optuna[5] used to tune the hyper-parameters Delayed ODE for modeling cytokine dynamics: Posterior parameter distributions for the ODE model: more efficient phagocytosis when guided by monocytes signaling Chemotaxis of neutrophils: • Based on ABM model • Molecular concentrations: PDE • Cell-molecules interactions t= 0 One ABM simulation takes approx. 16.5 mins 1 of blood contains: μL • 1000 pathogens • 500 monocytes • 5000 neutrophils Migration parameters for immune cells & are unknown MRM MRN ? MRM MRN dC(t) dt =s⋅Na(t)−kdeg ⋅C(t) dNa(t) dt =dN(t−d) dt −β⋅Na(t) secretion rate degradation rate delay between phagocytosis and secretion exhaustion rate concentration # active cells that secrete Used in hABM for hypotheses testing Initial condition for pathogens: ! • Novel quantitative and predictive simulation tool in infection research modeling •Multi-model approach: combination of different modeling techniques of increasing complexity within a single framework • Integration of different experimental data types: flow cytometric analyses, pathogen survival assay, proteomics data, microscopy data • Application to difficult-to-treat infections for predicting efficient treatment strategies C. albicans: + Neutrophils + Monocytes alive C. albicans: + Neutrophils + Monocytes alive ϕN [min−1] ρ [min−1] κN [min−1] κM [min−1] κEK [min−1] ϕM [min−1] γ [min−1] Posterior distributions for migration parameters: s [min−1] kdeg [min−1] β [min−1] d [min] 357 46 secretion rate [mins-1] 4 6 2 0 Probability density x104 x104 degradation rate [min-1] 0246x10-5 x104 Probability density 8 12 4 0 100 150 50 0x10-3 exhaustion rate [min-1] 2 6 10 14 Probability density 0.08 0.12 0.04 0 delay [min] 0510 15 Probability density MRM [μm2/min] Chemotaxis vs. random walk: Statebased model Hybrid agent-based model MuMoSim Transcription Cytokine release (TNF- , IL-6, ENA-78) α MRN [μm2/min]