Live demo
Biomarker screen
Across every biomarker, where is the signal? One row per biomarker for a comparison chosen once, a standardised difference between two groups or a correlation with one variable, with R's estimate and its interval on one axis without units, and R's p-values beside it, unadjusted and adjusted across the rows by Benjamini-Hochberg or Holm. The rows are sorted by the estimate, and a click on a row opens that biomarker in the group comparison or the association scatter, with a way back. The data is the synthetic study: 200 made-up participants, and no real one.
- The chart opens on the difference the synthetic study was planted with: every biomarker's change from Baseline to Week 4, Placebo against Treatment. Each row is one biomarker: R's standardised difference (Hedges' g, the first group's mean less the second's, in pooled standard deviations) as a dot, its 95% confidence interval as a line, on one axis without units with nought marked. IL-6 was planted with a difference between the arms; the others differ by chance alone.
- Beside each row are R's p-values: Welch's t-test, unadjusted, and adjusted across every biomarker that has one, by Benjamini-Hochberg or, under Statistics, Holm. The caption says how many tests the adjustment covered. Every result is exploratory. R is started in this browser when the page opens: the line under the screen says it is waiting, and what that first start downloads, until R answers. Nothing leaves this machine, and the chart computes no estimate, interval, p-value or adjustment itself.
- Click a row, or press Enter on it, to open that biomarker in the group comparison, in place, at the same visit and value with the same two groups and Welch's test; Back to the biomarker screen returns to the rows as they were. The filters go with you.
- Under Screen choose Correlation with one variable to screen every biomarker against a participant-level number, such as age, or against another biomarker at a visit: IL-10 at Baseline is the one the study was planted with, and TNF-alpha is then the top row. A row of a correlation opens the association scatter.
- Sort the rows by R's estimate, largest first, by name, or by R's adjusted p-value. With more biomarkers than fit, the rows are paged, twenty to a page. A row R could not compute shows R's reason and no number, and is left out of the adjustment, as R says. On a phone each row stacks: the biomarker, its interval, and the numbers beneath.
The controls, the filters, the listing, the participant profile and the Chart.js this chart draws with are safety.viz's, version 1.9.0, loaded beside bio.viz from a copy with a record of where it came from.
The statistics are gsm.bio's, version 0.1.0: R in this browser is given one file of R, copied from gsm.bio at commit 514cbc3 with a record of where it came from. The same file, run in desktop R on the rows this chart hands over, wrote the answers the browser tests hold this page to.