A discussion of tools and resources for mechanistic within-host modeling

Slides: https://www.andreashandel.com/presentations/

Andreas Handel

5/17/23

Acknowledgements

  • Lot’s of people contributed. See the websites for each tool/resource.
  • NIH funding.

Format for this presentation

  • I’ll start with a very short introduction to the topic.
  • The bulk will be show & tell demonstrations.
  • I’ll focus on tools/resources from our group.
  • There should be lots of time for questions and discussion.

Phenomenological/non-mechanistic/(statistical) models

  • Most widely used types of models, useful (almost) everywhere one has data
  • Look for patterns in data
  • Do not describe mechanisms leading to the observed outcomes (data)

\[ y = b_0 + b_1 x_1 + b_2 x_2 + ... \]

Mechanistic/process/simulation models

  • Are widely used in all areas of science.
  • Are simplified representations of specific processes/mechanisms.
  • Can be implemented as various different types of mathematical/computer models.
  • Can be used without and with data (and then also become statistical)

\[ \begin{aligned} \textrm{Bacteria} \qquad \dot{B} & = g B(1-\frac{B}{B_{max}}) - d_B B - kBI\\ \textrm{Immune Response} \qquad \dot{I} & = r BI - d_I I \end{aligned} \]

Simulation model uses

  • Explore the behavior of a system.
  • Make predictions about the behavior of a system.
  • With data: Perform inference, test hypotheses (doing stats).

Exploring/predicting cytokine-based interventions for TB (Wigginton and Kirschner, 2001 J Imm)

Simulation model types

  • Compartmental models are the simplest and most widely used.
  • Most commonly implemented as ordinary differential equations (ODEs).

Simple 2-compartmental model. (Handel et al 2020 Nat Rev Imm).

\[ \begin{aligned} \dot{B} & = g B(1-\frac{B}{B_{max}}) - d_B B - kBI\\ \dot{I} & = r BI - d_I I \end{aligned} \]

Simulation model types

  • Agent-based or network models are more detailed/complex.
  • Can be more realistic, but are more data-hungry and computationally demanding.

Acute virus infection. (Handel et al 2009 J Roy Soc Interface)

Questions/Discussion

  • This was a minimal background section to set the stage.
  • See Handel et al 2020 Nat Rev Imm for some more in-depth introduction.
  • I plan to now show various resources/tools for mechanistic/simulation modeling.
  • Any questions/feedback/comments before we move on?

Source: https://phdcomics.com/

Simulation Modeling in Immunology (SMI)

SMI Strengths/Weaknesses

  • The good
    • Decent amount of introductory material (videos, slides, readings)
    • Immunology and modeling
    • All free
  • The bad
    • No guided/curated experience
    • Coverage of topics is uneven/idiosyncratic
    • Not fully polished

SMI Questions/Discussion

Source: https://phdcomics.com/

DSAIRM R package

  • R package to learn/explore mechanistic simulation models: https://ahgroup.github.io/DSAIRM/
  • Modular design, starts with no-code approach, allows advancement to coding.
  • Each simulation app comes with model description and a set of suggested tasks that teach a specific modeling concept.
  • Full solutions to all tasks are available (on request).

Learning DSAIRM

DSAIRM Strengths/Weaknesses

  • The good
    • Starts out completely code-free.
    • Good number of within-host modeling topics are covered.
    • Set of guided exercises/tasks to help learn the material.
    • Full access to underlying model code.
  • The bad
    • Only the models I wrote are available.
    • Custom models (e.g., for research) will require adapting one of the existing models, which means having to write code.
    • Package name is maybe awkward (taking suggestions for better naming).

DSAIRM Questions/Discussion

Source: https://phdcomics.com/

modelbuilder R package

modelbuilder Strengths/Weaknesses

  • The good
    • You can build your own models (or adapt existing ones).
    • You can explore your models through the graphical interface.
    • You can get different versions of the model code for further use.
    • Has lots of overall potential 😄.
  • The bad
    • User interface not fully polished and robust.
    • No option to load and fit data (do stats).
    • Not suitable for big models.
    • Potential only partially realized 😄.

modelbuilder Questions/Discussion

Source: https://phdcomics.com/

flowdiagramr R package

flowdiagramr Strengths/Weaknesses

  • The good
    • Helps with potentially annoying model drawing task.
    • Is highly configurable and easy to use.
  • The bad
    • Currently can’t handle more complex models (e.g. 3-way interaction).
    • Still not fully tested, might still contain a good bit of bugs.

flowdiagramr Questions/Discussion

Source: https://phdcomics.com/

Summary and more discussion