Overview {-}

• This repository contains RMarkdown documents to reproduce the results figures for a research publication. Each figure is self-contained with explicit inputs and library imports by using RMarkdown's knit and merge approach¹.

• All extended data and supplementary figures are created alongside the results figures. The workflow overview in extended data 1 is not present here because it was created using BioRender.com.

• Download the input data².

• To recreate the figures code, if necessary, edit the config.R file in the R subdirectory with the path to your downloaded data directory, then run these R commands to install the dependencies and then build the RMarkdown book:

  ## Install the dependencies:
  install.packages("pak")
  pak::local_install_deps(dependencies = TRUE)
  
  ## Build the RMarkdown book:
  bookdown::render_book()

  The above workflow has been tested on Ubuntu 24.10--26.04 GNU/Linux, Windows 11, and macOS 15.3--26.7

• An additional download contains the job submission scripts that generate the input data. However, these scripts are fairly specific to the compute clusters³ on which they were run and require over a quarter million compute hours to complete and are, therefore, intended as reference rather than being part of a practical, fully-reproducible workflow.

• If you need to cite this work, see the CITATION.cff and the associated preprint:

  @article {Nanda2026.10.09.758083,
  	author = {Nanda, Pariksheet and McCaffrey, Erin F. and Michael, Christian T. and Delmastro, Alea C. and Michael, Gabrielle and Mattila, Joshua T. and Flynn, JoAnne L. and Angelo, R. Michael and Kirschner, Denise},
  	title = {A tissue time machine to forward-predict single-timepoint data applied to tuberculosis granulomas},
  	elocation-id = {2026.10.09.758083},
  	year = {2026},
  	doi = {10.64898/2026.10.09.758083},
  	publisher = {Cold Spring Harbor Laboratory},
  	abstract = {Recent advances have improved understanding of infection; however many detailed datasets are based on single timepoints post-infection and cannot generate longitudinal insights needed to predict infection outcomes. For example, pulmonary infection with Mycobacterium tuberculosis (Mtb) initiates a complex system of immune and bacterial interactions that leads to granuloma formation, yet single snapshots in time need elaboration long-term. We develop a new approach, the tissue time machine, applied to predict outcomes of non-human primate granulomas from initiation to endpoint using single timepoint tissue datasets including high-dimensional spatial features. This involves identifying granuloma digital partners, simulations most closely resembling real granulomas from a virtual library of over 50,000. By analyzing digital partners, we predict infection outcomes of 22/28 granulomas, and that Mtb-specific, IFN-γ-producing T cell recruitment rates influence uncertainty for the 6/28. This method is applicable to any system that allows for mechanistic predictions over time using heterogeneous, incomplete, and high-dimensional datasets.Competing Interest StatementThe authors have declared no competing interest.NIH Common Fund, https://ror.org/001d55x84, R01 AI50684, F31 AI191425, NIH S10 OD026825Wellcome Leap Delta Tissue Program},
  	URL = {https://www.biorxiv.org/content/early/2026/10/12/2026.10.09.758083},
  	eprint = {https://www.biorxiv.org/content/early/2026/10/12/2026.10.09.758083.full.pdf},
  	journal = {bioRxiv}
  }

Footnotes

1. Knit then merge (K-M) approach https://bookdown.org/yihui/bookdown/new-session.html ↩

2. Simulation and processed data input for publication figures generated by the spatialmatch R package https://zenodo.org/records/15102620 ↩

3. Purdue University's Anvil cluster (primarily), San Diego Supercomputer Center's Expanse cluster, and the University of Michigan's Lighthouse and Great Lakes clusters (these latter non-Purdue clusters were only used to split up running the 60,000 GranSim simulations). ↩