stochasticfinitemodel

R software for fitting and simulating the stochastic finite (SF) model (Pridham and Rutenberg 2023).

Getting started

Download the latest release (tags, right hand side of screen). You can open sf.R in R and start fitting data now. I recommend you read the vignettes which are written in R notebook (.Rmd) using RStudio. (You can read the .html instead but it is less convenient.)

You should start with sfvignettefitsf.Rmd. You will need to source sf.R. You'll also need to enter the location of this repository's contents, which is stored in outputDir.

What it does

The SF model uses longitudinal data to estimate an interaction network and equilibrium (steady-state) behaviour. It is a linear approximation of a more generate stochastic (Wiener) process model.

Requirements

Written for R version 4.1.1. RStudio is recommended for the vignettes.

Required packages: MASS and survival.

Recommended: parallel

Optional: nloptr.

How to use

See vignettes for functionality. Vignettes are written as Rstudio notebooks (version 2023-03-00) (.Rmd).

The .html are included for completeness but have not been optimized (so they may be ugly or lack figures).

Key file

sf.R is main file which contains the functions needed to fit the model.

Tips

1. Although not the default, it is worth trying the linear regression estimator: FitFun=FitLM.SF e.g. sf = FitSF(y=y,x=x,dt=dt,FitFun=FitLM.SF). The linear regression was proven in a subsequent paper, and usually works a lot better.
2. If you have reduced rank W the model will likely fail to fit, this is due to collinearity problems. You can (a) reduce the parameters used in the fit or (b) pre-process using principal component analysis e.g. sf = FitSF(y=y,x=x,dt=dt,Npc=ncol(y)-2) will drop the last 2 principal components, thus reducing collinearity. You can also try fitting a diagonal model, this works pretty well if you preprocess with principal component analysis e.g. sf = FitSF(y=y,x=x,dt=dt,Npc=ncol(y),diagonalW=T).

How does the math work?

See Pridham and Rutenberg (2023). The supplemental is quite detailed.

Cite as

Pridham, G. & Rutenberg, A. D. Network dynamical stability analysis reveals key ‘mallostatic’ natural variables that erode homeostasis and drive age-related decline of health. Sci. Rep. 13, 1–12 (2023). https://doi.org/10.1038/s41598-023-49129-7

Bibtex

@ARTICLE{Pridham2023-pi, title = "Network dynamical stability analysis reveals key ``mallostatic'' natural variables that erode homeostasis and drive age-related decline of health", author = "Pridham, Glen and Rutenberg, Andrew D", journal = "Sci. Rep.", publisher = "Nature Publishing Group", volume = 13, number = 1, pages = "1--12", month = dec, year = 2023, url = "https://www.nature.com/articles/s41598-023-49129-7", language = "en", issn = "2045-2322, 2045-2322", doi = "10.1038/s41598-023-49129-7" }

See also

It is wise to use the least-squares estimator which is proven in this paper:

Pridham, G. & Rutenberg, A. D. Dynamical network stability analysis of multiple biological ages provides a framework for understanding the aging process. J. Gerontol. A Biol. Sci. Med. Sci. (2024) doi:10.1093/gerona/glae021

Bibtex

@ARTICLE{Pridham2024-cy, title = "Dynamical network stability analysis of multiple biological ages provides a framework for understanding the aging process", author = "Pridham, Glen and Rutenberg, Andrew D", journal = "J. Gerontol. A Biol. Sci. Med. Sci.", month = jan, year = 2024, url = "http://dx.doi.org/10.1093/gerona/glae021", keywords = "biological age; complexity; eigen analysis; systems biology", language = "en", issn = "1079-5006, 1758-535X", pmid = "38206765", doi = "10.1093/gerona/glae021" }

An application that helps explain how the model works: Pridham, G., Tennankore, K. K., Rockwood, K., Worthen, G. & Rutenberg, A. D. Systems-level health of patients living with end-stage kidney disease using standard lab values. arXiv q-bio.QM

@ARTICLE{Pridham2024-ql, title = "Systems-level health of patients living with end-stage kidney disease using standard lab values", author = "Pridham, Glen and Tennankore, Karthik K and Rockwood, Kenneth and Worthen, George and Rutenberg, Andrew D", month = may, year = 2024, url = "http://arxiv.org/abs/2405.20523", archivePrefix = "arXiv", eprint = "2405.20523", primaryClass = "q-bio.QM", arxivid = "2405.20523" }

Common errors

-Individuals are sorted by their age, so make sure age captures changes in time e.g. if you measure someone at age 20 then 3 months later then the followup age should be 20.25 (NOT 20!) -If you use Reshape() with an id column that can't be coerced into numeric, it will crash. I'll remove this in the future. -If you use Reshape() with character sex it will fail to perform sex-specific normalization. In this case you should binarize sex or, (i) set sexSpecific=FALSE or (ii) set preprocess=FALSE. -namespace issues might also crop up (for me it's Impute), you can always add a post hoc wrapper for this or just load the package last.