95% Confidence Interval Calculation in Statistical Analysis [one factor] Scores Plots

[jessie_olough]

Hello,

Would it be possible to know which statistical method is used to calculate the 95% confidence intervals in the PCA and PLS-DA scores plots in the Statistical Analysis [one factor] section of the MetaboAnalyst website, please?

I found another response on this forum for how 95% confidence intervals are calculated for the biomarker analysis (link to it below). Is this the same method for the Statistical Analysis [one factor]?

How does MetaboAnalyst calculate 95% confidence intervals (CIs) of AUC/pAUC (biomarker analysis)?

The program uses a non-parametric re-sampling based approach for calculating the confidence intervals. In either bootstrap (classical) or MCCV (multivariate) approach, the data are re-sampled many times, with the AUC/pAUC calculated each time. Confidence intervals are then estimated by simply sorting the data and taking the percentiles to our desired confidence interval bounds. For instance, in 1000 resampling, 95% CIs will be the 25th and 975th values of the sorted AUC values. For details, ple…

I am trying to replicate these plots in Python, but I am finding that using either the standard t- or normal- distribution methods gives far smaller confidence regions than for corresponding plots in MetaboAnalyst.

How to Calculate Confidence Intervals in Python

Any help would be much appreciated.

Kind regards,
Jessica

[jeff.xia]

You can always find out the underlying R code in our GitHub, see below on computing the data points.

…
groupVar<-var(cbind(pc1[inx],pc2[inx]), na.rm=T);
groupMean<-cbind(mean(pc1[inx], na.rm=T),mean(pc2[inx], na.rm=T));
pts.array[,i] ← ellipse::ellipse(groupVar, centre = groupMean, level = reg, npoints=100); # reg is confidence level
…

The ellipse R package: CRAN: Package ellipse