afm-analysis / nanite Goto Github PK

• ancillary parameters are better than lmfit expressions

Sometimes the approach curve (before indentation) exhibits a non-zero slope. While many might be of the opinion that such curves should be discarded altogether, sometimes there is just not enough data of better quality to analyze.

I believe the question is not whether we should avoid such an analysis (because the need is there). But how a solution could look like. There are two approaches with benefits and drawbacks:

1. Measure the slope and subtract it from the entire approach part (including the indentation part). Here we are effectively manipulating measurement data, which is not so good. In addition we would have to make sure that approach and retract curve "match up" after subtraction of the approach curve. This results in an offset of the approach curve and might also affect hysteresis.

2. Fit the slope and subtract it only up until the fittted contact point. In contrast to point 1, this is not anymore a preprocessing step, but part of the fit model. Here we are not touching the indentation data, which might be better.

Personally, I would prefer approach 2. Open for discussion. Please attach your data as .zip files to this issue if you would like to contribute to the discussion.

Create a new Python package called afmformats.

• create a transparent standard class for data file formats
• each file format implements a new subclass
• data files should not be loaded into memory (memory consumption is too large for qmaps); either load data lazily or cache them on disk using hdf5

AFM-analysis/PyJibe#7

• Check that the parameter keys in a model are reflected (order, name) in the arguments of the modeling function
• Make sure that length of parameters, units, etc. lists are the same.

...similar to preprocessing decorators. This will make the code a lot cleaner.

...to create memory-efficient copies of Indentation datasets. Useful if the user wants to play with models/parameters without changing the current fit properties.

something about array sizes

This feature was introduced with ca51c57

It might be a good idea to use this as the default contact point estimator. It is quite robust.

Pros:

• seems like it is really robust

Cons:

• Might be slower (This needs to be checked)

I.e. data are recorded at 1kHz sampling rate, but fits should be performed at 100 Hz sampling rate. See e.g. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7610513/

Currently this is not possible

• Which one works best for which data?
• example matrix (force-curves vs. CP estimation algorithm)
• Add a new section in the docs

The minimisation method of Levenberg - Marquardt doesn't work well for exponential decays.
the KVM new model function is being fitted with Nedler (xtol and fool = 1e-20) method on the specific residuals defined in the model function.

Compare with data from previous publications (e.g. https://dx.doi.org/10.3390/polym10101055).
original paper: https://doi.org/10.1098/rspa.1971.0141
visualization: https://doi.org/10.1016/j.colsurfb.2015.06.044

Implement a rating refiner (user sees two curves and decides which one is better). The model is updated at each iteration and the largest deviations are presented to the user.

• support floats in training step in nanite
• implement rating refiner

When the maximum adhesive (pull-off) force F ad is large enough in
comparison with the maximum force F max applied during load-
ing, i.e. for |F ad /F max |> 0.05 [28], which is frequently observed
in retraction curves, Hertzian analysis is no longer valid and
models for adhesive contact of spheres should be invoked. These models
include the JKR (Johnson–Kendall–Roberts), DMT
(Derjaguin–Muller–Toporov), and Maugis models

The dimensionless Tabor parameter � = (R� 2 /E 2 � 3 ) 1/3 is
used to choose the suitable model [31]

"D. Tabor, Surface forces and surface interactions, J. Colloid Interface Sci. 58 (1977) 2–13."

nanite should be able to fit the approach and retract segments of creep-compliance data.

• This requires some refactoring and abstraction. How do we deal with the afmformats.AFMData class?
• Also related to AFM-analysis/afmformats#15.
• This should also work for stress-relaxation and others (they are all segment-based, so it should be trivial)

Hello,
Pyjibe sometimes fits a flat line (see figure) even though the contact point found is out of the range of tip position. My guess is that this is connected to the negative right range chosen by the user (because the contact point wasn't found correctly in the preprosessing, I had to choose a smaller range)
could you take a look at this?

This should be part of the "baseline correction" preprocessing step with the default being "offset".

• use line and polynomial fit to determine slope
• #7
• options: "subtract up until contact point", "subtract from approach part", or "subtract from entire dataset".
  The user has to decide which makes sense.