Peak finding functions are certainly not the end all and be all of assessing peaks in AFM images of surfactant protein D. I am totally disappointed in how they work.  It is so clear that depending upon measures that have preceeded a peak and how big a peak is has LITTLE at all to do with what nature does with proteins (specifically, irregularly placed bumps and twists of varying sizes that can be seen by eye with electron microscopy and AFM and other microscopic media). Not a single peak finding function that I have found has any real comparison to what a trained eye can find (at least at this time). Since the user determines the robotic approach, peaks can vary from 4 to 40 when the eye can see about 15 in a dodecamer regularly.

When I see the divisions of peaks, the numbered peaks, the missed peaks that signal processing produces i think to myself. Why that, or why not this.

Sadly this makes everything I have tried to figure out about unbiased counting, height, width and valley depths, using image processing and peak finding, in surfactant protein D trimers (and dodecamers) is pretty much NOT verifying anything except the fact that they dont work.

Being adaptable mentally just is still a human trait……. how much longer, I dont know. Training seems still to lag behind human abilities.

Things I have noticed:

1– in the trimers, it appears that multiple peaks clustered (maybe 2 or 3) seem to happen when there are (as in the case of the glycosylation site (published data) the three trimers at the glycosylation site have shallow staggered peaks, as if there is a rotary positioning. Seems logical that the lumpy effect would be found when rotating a repeating event (like glycosylation).
2– signal processing functions are not very adaptable, they do not see pattern well, they ignore some peaks and find others in a way that is inflexible

3– N term peak height in the dodecamers and higher multimers shows variation in N term attachments.