Back

Help on options

Estimated time:
This icon indicates a rough estimate of the time that will be needed to perform all the panel computations (including cross-validation and pre-filtering if those are selected). This time is subject to many variations (such as server load, or some characteristics of the data) but should give a good idea of the order of magnitude of computation time (second, minutes, hours, days or years).
Maximum panel size
This number is the maximum size you want your panels to be. For instance, to test for panels containing 1, 2, 3 or 4 biomarkers, set this list to 4.
Minimum
  • Specificity or sensitivity : the focus of your panel: do you want to have a high specificity or sensitivity ?
  • Number: this number defines the minimum specificity or sensitivity (depending on your choice above) that must be achieved by a panel to be retained. All panels that doesn't reach this performance will be discarded.
For instance if you choose specificity and 0.95 (the default), PanelomiX will optimize the sensitivity of panels with at least 95% of specificity.

Cross-validation

Cross-validation (CV) is used to assess the stability of panels. Choose yes to enable cross-validation (enabled by default). More details about cross-validation on wikipedia.

Cross-validation folds
Determines in how many parts the data will be divided for CV.
Cross-validation repeats
Cross-validation can be repeated to get better estimates of the stability, but more repeats mean longer computation time.

Pre-filtering

Pre-filtering is sometimes required to speed-up the computation. Choose yes to enable pre-filtering (disabled by default).

If enabled, a Random forest classifier is trained (with R's randomForest package) and the following filters can be selected:

Number of markers
Only this number of markers will be used in the computations. If more markers are selected in the details, only those being selected as split points most often will be retained.
Number of thresholds
Several thresholds are tested for each markers. By default, these thresholds are the ROC curve local maximas. If an integer N is defined here, only the N thresholds being selected most often as split points for the given marker will be retained.

Output

The output options define how the results will be presented.

Desired output format
With “HTML only”, a simple HTML output is produced. This is the fastest option, best used during early screening steps.
If “PDF report” is selected, a PDF file is generated summarizing the results (including cross-validation if this option was selected above). A ROC analysis is performed without computing confidence intervals.
When “PDF report with confidence intervals” is checked, the PDF report is generated as above, but ROC curves come with additional confidence intervals for (p)AUC, sensitivity and specificity. This is most complete but also the slowest option.
Skip cross-validation of individual biomarker
Check this box only if running the panel with cross-validation results in an error message saying “Cannot cross-validate the best biomarker. Most likely, its performance is too low. Try again and if the problem persists, disable cross-validation.” This message means that the best biomarker could not meet the performance requirements defined above in the “Minimum Accuracy/Sensitivity/Specificity Number”. In this case, no panel is found and the cross-validation cannot succeed, causing the PDF export to abort. With this option checked, this part of the cross-validation will be ignored and the PDF export will succeed.