Skip to content
Sufyan Suleman

ISAT

Insulin sensitivity, secretion and glucagon indices in the browser or from the command line, no installation

Launch browserSource on GitHubCite
WEB APP AND CLIISAT47 published metabolic indices,computed in the browserFor researchers with cohortor trial measurementssufyansuleman.github.io0306090120

Overview

ISAT, the Insulin Sensitivity Analysis Tool, computes 47 published surrogate indices from fasting, OGTT, lipid and anthropometric measurements, for a whole research file or for one person: 33 indices of insulin sensitivity and resistance, and 14 of insulin secretion and glucagon. It runs as a web application in the browser with nothing to install and no account, and as a single-file Linux command-line tool for larger files and cluster work. The two share one calculation engine, so their results are identical.

Research problem

The published indices each come with their own formula, required timepoints and unit conventions, and re-implementing them for every study is a recurring source of error. Many researchers who need these indices do not write code, and those who do often need to run them on files too large for a browser. ISAT covers both without changing the calculation.

In the browser

  • 33 insulin sensitivity and resistance indices, including HOMA-IR, QUICKI, Matsuda, Stumvoll, Gutt, BIGTT-SI, TyG, VAI and Adipo-IR
  • 14 beta-cell and alpha-cell indices, including HOMA-beta, the insulinogenic index, the corrected insulin response, the disposition index, BIGTT-AIR and glucagon AUC, following IsletCalc and Madsen et al. 2024
  • Optional glucagon columns at 0, 30, 60 and 120 minutes, in pmol/L or pg/mL, unlock the glucagon indices
  • CSV or TSV upload for up to 100,000 individuals, with warnings when values do not fit the chosen units
  • Results explored as per-index summaries, distributions by sex, Spearman correlation heatmaps and scatter plots
  • Transformation for downstream analysis by log, z-score or rank-based inverse normal, optionally within sex
  • Download of results as CSV, a settings file for reproducibility, and every plot as SVG
  • Each index shows its formula, its reference, the population it was derived in and how the formula was verified

Everything is computed locally in the browser and no data are uploaded.

On the command line

The Linux build is a single file that needs no installation, no root access and no R, Python or Node runtime. It is intended for files beyond the browser limit and for HPC use.

# download the latest release
url=https://github.com/sufyansuleman/isat/releases/latest/download
curl -L $url/isat-linux-x64.tar.gz | tar xz

./isat check cohort.csv                       # columns, units, problems
./isat calculate cohort.csv -o results.csv    # all indices

# or a subset of index groups
./isat calculate cohort.csv -o results.csv --include fasting,secretion

# or in chunks, for example a SLURM array
./isat calculate cohort.csv -o res_$i.csv --chunk $i/20
./isat merge res_*.csv -o results.csv
./isat transform results.csv -o rint.csv   --method rint --within-sex --input cohort.csv

Builds are available for x64 and ARM, with checksums, alongside a Node version of the same tool.

Validation

Results are validated against InsuSensCalc, an independent re-implementation of the same formulas in R, and against base R for all transformations and statistics. Deliberate differences between the two are documented in the repository.

These are surrogate indices, not direct measurements of insulin sensitivity, and not a diagnosis.

Derivation populations

Each index was developed in one population, mostly single European or United States cohorts of a few hundred people or fewer, and most of the original papers do not report ancestry. The repository lists the population, sample size, place and original paper for every index. Values are best compared within a study rather than against thresholds carried over from another population.



Do you need insulin sensitivity indices without writing any code?

ISAT runs in the browser for files up to 100,000 people and as a single Linux binary for anything larger. I can help you map your columns, check units and set it up on your cluster.

Get in touch