Overview: Computational models of disease progression have been constructed for a myriad of pathologies. Until now, the conceptual implementation for pathology-related in silico intervention studies has been ad-hoc and similar in design to experimental studies. We introduce the multi-scale interventional design (MID) framework toward two key goals: tracking of disease dynamics from within-patient to patient to population scale; and tracking impact(s) of interventions across spatial scales.

The MID framework prioritizes investigation of impact on individual patients within virtual pre-clinical trials, instead of replicating the design of experimental studies. We use HostSim: our next-generation whole patient-scale computational model of individuals infected with Mycobacterium tuberculosis. HostSim captures infection within lungs by tracking multiple granulomas, together with dynamics occurring with blood and lymph node compartments. We extend HostSim to include simple drug intervention and use the virtual cohort framework to quantify the impact of interventions at the granuloma (tissue), patient (lungs, lymph nodes and blood), and population scales. Sensitivity analyses allow us to determine which features of virtual patients have the strongest impact on intervention efficacy across scales. Coupling all of this with the MID approach allows us to identify mechanisms principally interacting with modes of interventions at multiple scales and to examine features driving patient-response heterogeneity.

MID requires three key components to be given in detail: (i) a collection of virtual patients, given along with a biological justification as to why the same virtual patient is able to be represented in multiple models; (ii) a set of two related and individually credible model versions, such as a control model and an experimental model if representing an experimental intervention; and (iii) an impact quantification method by which the outcomes of both model versions can be meaningfully compared.

Infection progression of TB in a virtual host both without (Left) and with (Right) drug intervention. A simplified representation of BDQ has been given to this virtual host on day 200. Each video shows a HostSim time-lapse video for 400 days showing virtual lungs, granulomas, lymph nodes, and blood cell concentrations, colored by the brightness (FDG avidity prediction) predicted from PET-CT scans. Blood is colored by concentration of Mtb-specific CD4⁺ T cells. The lungs and body of the virtual NHP are triangulations of digitized lung and NHP body surfaces provided by the lab of JoAnne Flynn. The size of each lymph node is determined by total number of T cells, and granulomas diameters are proportional to the predicted diameter based on cell-count and volume of caseum - the necrotic core present in many granulomas.

Impact quantification of BDQ-like drug intervention applied on day 200. Granuloma impact score (GIS) is a log-ratio of bacterial load with to without drug intervention for each granuloma. Some granulomas (blue) sterilized with and without drugs by day 400, though earlier sterilization occurs. Other granulomas (green) sterilize only with the addition of drugs, while others still (black) sterilize in neither.

This approach allows us to closely examine the parameters that determine how effective the drug will be for every host. The key to this is to separate parameters that are intrinsic to the virtual host's identity (VH) and those parameters that are intrinsic to the definition of the intervention. In this way, we use the same VH as inputs to both the control version and intervention version of the model to obtain a composite function , to which we can apply traditional sensitivity analyses, such as PRCC.