cremerlab/hplc-py

A Python utility for the processing and quantification of chromatography data

hplc-py's Issues

Help with peak fitting parameters

Hello, I'm having some difficulty understanding the hplc-py peak fitting parameters and tuning them appropriately. It could be helpful to offer more examples of how changing them affects the results.

What parameters would you suggest to integrate this chromatogram?
Default:

Using approx_peak_width=0.25 (uncommented the line in the above codeblock):

Neither of the two examples seems to be visually valid, and I'm not finding a combination of parameters that would make sense.

For reference, this is what Chemstation returns:

Setting `fit_peaks` prominence below 0.2 results in upper bounds with lower value than lower bounds.

I am testing the application of fit_peaks to my dataset, whose profile you can see below:

The behavior is as expected unless I set peak prominence below 0.2, at which point I receive an error from least_squares informing me that at least one upper bound is less than the corresponding lower bound.

The error is below:

Traceback (most recent call last):
  File "/Users/jonathan/mres_thesis/wine_analysis_hplc_uv/src/wine_analysis_hplc_uv/dataset_eda/peak_analysis.py", line 125, in <module>
    main()
  File "/Users/jonathan/mres_thesis/wine_analysis_hplc_uv/src/wine_analysis_hplc_uv/dataset_eda/peak_analysis.py", line 122, in main
    HPLCPy(data)
  File "/Users/jonathan/mres_thesis/wine_analysis_hplc_uv/src/wine_analysis_hplc_uv/dataset_eda/peak_analysis.py", line 103, in __init__
    peaks = chm.fit_peaks(
            ^^^^^^^^^^^^^^
  File "/Users/jonathan/Library/Caches/pypoetry/virtualenvs/wine-analysis-hplc-uv-F-SbhWjO-py3.11/lib/python3.11/site-packages/hplc/quant.py", line 801, in fit_peaks
    peak_props = self.deconvolve_peaks(verbose=verbose,
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/jonathan/Library/Caches/pypoetry/virtualenvs/wine-analysis-hplc-uv-F-SbhWjO-py3.11/lib/python3.11/site-packages/hplc/quant.py", line 671, in deconvolve_peaks
    popt, _ = scipy.optimize.curve_fit(self._fit_skewnorms, v['time_range'],
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/jonathan/Library/Caches/pypoetry/virtualenvs/wine-analysis-hplc-uv-F-SbhWjO-py3.11/lib/python3.11/site-packages/scipy/optimize/_minpack_py.py", line 974, in curve_fit
    res = least_squares(func, p0, jac=jac, bounds=bounds, method=method,
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/jonathan/Library/Caches/pypoetry/virtualenvs/wine-analysis-hplc-uv-F-SbhWjO-py3.11/lib/python3.11/site-packages/scipy/optimize/_lsq/least_squares.py", line 814, in least_squares
    raise ValueError("Each lower bound must be strictly less than each "
ValueError: Each lower bound must be strictly less than each upper bound.

The bounds in question are below, where 'lb' is lower bound, 'ub' is upper bound:

	lb	ub
0	0.000378979	0.0378979
1	0	0
2	0.0004896	0
3	-inf	inf

As I understand it, the prominence is the normalized prominence, i.e. the percentage of the maxima as a threshold for inclusion of a given peak. For the same dataset i've been using a value of 0.05 to ignore the smallest peaks with no problem.

Replace SNIP implementation with broad integration of `pybaselines`

I've developed hplc-py using only one baseline estimation algorithm (SNIP), but there are many options that can be used. These are nicely tested and developed in the pybaselines package. A development goal of hplc-py would be to deprecate the correct_baseline method of a Chromatogram with a call to the broad pybaselines API.

Small issue in correct_baseline method

Hi, cool library! I noticed a small issue in the correct_baseline method in quant.py.

hplc-py/hplc/quant.py

Lines 884 to 885 in 8ce5223

	tform_new[j] = min(tform_new[j], 0.5 *
	(tform_new[j+i] + tform_new[j-i]))

Within the inner loop, the values of tform need to be used for the minimum comparison rather than tform_new. Since tform_new is being written to within the loop, using its values for the comparison will cause some unforseen issues. To help compare with Morhác's original SNIP implementation, the working vector (w in the paper) is tform_new and the current vector (v in the paper) is tform.

The suggested change is show below:

for i in iter:
    tform_new = tform.copy()
    for j in range(i, len(tform) - i):
        tform_new[j] = min(tform[j], 0.5 * (tform[j+i] + tform[j-i]))  # changed here
    tform = tform_new