A low assumptions approach to investigating results

When you reach the review stage of the workflow, you are presented with a list of all the detected features along with their anova p-value and maximum fold change between your groups.

Instead of going through the list and reviewing each feature individually, it would be a good idea to do a quick quality check and get an overview of your data. To do this, we take all of the detected features (whether ranked significant or not), and perform Principal Component Analysis. Obviously this may include non-spot material, such as streaks, but it can be useful to consider this first as well.

To do this, select all the detected features in the list by clicking anywhere inside the list and pressing Ctrl + A to select all rows. Then click the "Include spots in results" button under the list to tick all the selected features. Once you have done that, click "Section Complete" to go to Progenesis Stats.

The statistics stage of the workflow will perform Principal Component Analysis on everything you have 'ticked' at the previous stage (in this case, all detected features) and display the results like this:

The grey ellipse in the centre is all of the spot numbers, you can ignore those for now. What we are interested in at this point is the pink and blue dots. Each coloured dot represents an image in this experiment. The pink dots are the control group and the blue dots are the treated group. PCA does not use any information about the groups when doing its calculations, so because we can see a clear separation between the groups here, the data has clear grouping characteristics.

One of the gels in the treated group (a blue dot) seems to be a bit of an outlier because it's much farther away from the other gels in the group. If you move the mouse over that blue dot, you'll see that it's gel "07-0244" so you may want to keep in mind that this gel appears to be behaving slightly differently from the rest of the group and possibly investigate what is different about this gel.

Next, we can sort by Anova p-value and tag all spots with a p-value < 0.05, then filter to use only these spots for the PCA.

Right click on a selection of spots to assign a tag

Enter the name for the new tag

Use the drop down menu on the tag column to filter the spots

This should give us a clearer separation between the groups because we're now selecting only those features that are changing significantly between the groups (i.e. with a p-value < 0.05). Any features that are not changing between the groups are ignored. The new PCA plot is:

The groups (i.e. lines of pink dots and blue dots) may look closer together, but the Principal Component 1 axis is now showing 65% of the variance in the selected data, whereas previously it was only showing 29% of the variance. So the groups are more strongly separated in this PCA plot.

As before, gel "07-244" is behaving differently from the rest of the treated group. However that difference is much smaller when looking at just the significantly changing features. You can see by looking at the amount of variance shown on each axis: 65% for Principal Component 1 and 10% for Principal Component 2.

Next, we can sort by power and tag all those features with a power > 0.8. This will give us features that have both an Anova p-value < 0.05 and Power > 0.8. Filtering on the new tag, as before, gives us 806 features in this dataset (from 3,316 detected features in total). The highest q-value for these features is 0.0176, which estimates that around 1.76% of these features (about 14 features) will be false positives.

Now that we've used the stats to filter down the number of spots we need to consider, we can go back to the review stage and tag each of those spots from our filtered list we want investigate further. E.g. spots that are suitable for picking.