Re: Deconvolution of 1D spectra

[Jonathan Helmus]

On 12/03/2014 04:39 AM, Kresten Bertelsen wrote:

Hi,

I'm stuggeling with fitting of 1D spectra using the fit_spectrum function.

 

Below is the code which gives the error and an overview of my parameters.

 

 

# Read in some data

 

import nmrglue as ng
import numpy as np

 

dic, data = ng.bruker.read_pdata('./Data Files/P31_1D/10/pdata/1')
SF=dic['procs']['SF']
SW_p=dic['procs']['SW_p']
SI=dic['procs']['SI']
OFFSET=dic['procs']['OFFSET']
AX=np.linspace(OFFSET,OFFSET-SW_p/SF,SI)

 

# Peak pick and deconvolute

 

peaktab = ng.analysis.peakpick.pick(data,np.max(data)*0.02)

# Prepare the inputs for deconvolution

lsh= list(["g" for x in range(np.size(peaktab['X_AXIS']))])

params=[]
for idx,peak  in enumerate(peaktab['X_AXIS']):
    params.append((peak,500))

amps=list(data[[int(x) for x in peaktab['X_AXIS']]])
bounds=[(0,1000) for x in range(np.size(peaktab['X_AXIS']))]
ampbounds =[(0,amps) for x in range(np.size(peaktab['X_AXIS']))]
centers = list(peaktab['X_AXIS'])
rIDs = list(tuple(peaktab['cID']))
box_width = tuple([(2000) for x in range(np.size(peaktab['X_AXIS']))])
a1, a2, a3, a4, a5 = ng.analysis.linesh.fit_spectrum(data,lsh,params,amps,bounds,ampbounds,centers,rIDs,box_width,0)

 

 

This code gives the following error

C:\Anaconda\lib\site-packages\nmrglue\analysis\linesh.py in fit_spectrum(spectrum, lineshapes, params, amps, bounds, ampbounds, centers, rIDs, box_width, error_flag, ve
rb, **kw)
    243 
    244         for i in range(len(shape)):
--> 245             bmin[:, i][np.where(bmin[:, i] < 0)] = 0
    246         for i, v in enumerate(shape):
    247             bmax[:, i][np.where(bmax[:, i] > v)] = v

IndexError: too many indices for array

 

Here is an overview of my variables

Variable    Type        Data/Info
---------------------------------
AX          ndarray     131072L: 131072 elems, type `float64`, 1048576 bytes (1 Mb)
OFFSET      float       24.89154
SF          float       242.937117583
SI          int         131072
SW_p        float       12175.3246753
ampbounds   list        n=11
amps        list        n=11
bounds      list        n=11
box_width   tuple       n=11
centers     list        n=11
data        ndarray     131072L: 131072 elems, type `int32`, 524288 bytes (512 kb)
dic         dict        n=1
idx         int         10
lsh         list        n=11
ng          module      <module 'nmrglue' from 'C<...>es\nmrglue\__init__.pyc'>
np          module      <module 'numpy' from 'C:\<...>ages\numpy\__init__.pyc'>
params      list        n=11
peak        float64     69279.0
peaktab     recarray    [(58584.0, 1, 34.0, 22646<...>021814548, -816324222.0)]
rIDs        list        n=11
x           int         10

 

--
You received this message because you are subscribed to the Google Groups "nmrglue-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to nmrglue-discu...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Kresten,


    The fit_spectrum function is very particular about the formatting of the parameters, some of the ones you are passing have the incorrect layout.  The documentation of the function [1] goes into detail on the expected layout of the various parameters.  For starters, the bounds parameter has a min and max value for only one lineshape parameter.  Gaussians lineshape's have two parameters, peak location and linewidth, so you must provide bounds for both of these parameters.  Also keep in mind that all the lineshape parameters are in units of points and not NMR units like Hz, ppm, etc.

    I've been meaning to write up an example of using fitting a spectrum using nmrglue's pick and fit_spectrum functions.  Give me a few days and I should be able to come up with something to add to the documentation.  I think having an example script that does something similar would be more helpful then running through the issues with the various function parameters.

    Cheers,

    - Jonathan Helmus


[1] http://nmrglue.readthedocs.org/en/latest/reference/generated/nmrglue.analysis.linesh.fit_spectrum.html#nmrglue.analysis.linesh.fit_spectrum

[Paul Driscoll]

Hi Jonathan,

Did this example ever appear? I'm currently trying to work out how to define bounds and ampbounds (1D spectrum at the moment)... Example code would be very helpful.

Cheers,

Paul

[Jonathan Helmus]

Paul,

    There is an example, "Find and fit peaks in a spectrum" in the nmrglue wiki that provides an example of using the fit_spectrum function in nmrglue.  https://github.com/jjhelmus/nmrglue/wiki/Find-and-fit-peaks-in-a-spectrum

Cheers,

    - Jonathan Helmus

[Paul Driscoll]

Great Thanks Jonathan!

[Alexandra Grigore]

Hi Jonathan,

I've been trying to run the fitting example in the wiki and I get the following error:

[....]

  File "C:\Users\nemes\AppData\Local\Programs\Python\Python35-32\lib\site-packages\nmrglue\analysis\linesh.py", line 277, in fit_spectrum

    campbounds, wmask, error_flag, **kw)

  File "C:\Users\nemes\AppData\Local\Programs\Python\Python35-32\lib\site-packages\nmrglue\analysis\linesh.py", line 459, in fit_NDregion

    if len(dim_bounds) != dim_nparam[j]:

TypeError: object of type 'zip' has no len()

Reading and plotting data works fine though.

Also, what I actually need is not fitting but calculating the FWHM for many spectra like the one attached (which is proccessed from Bruker data with nmrglue), basically what peakw function does in TopSpin. I wanted to first fit a gaussian to the peak and then calculate it manuall. However, I see there is also this function: nmrglue.analysis.lineshapes1d.center_fwhm(signal) - but I am not sure how to use it. I'd really appreciate any hints. I tried using several general python scripts I found online for calculating the FWHM, but the results were pretty rubbish...

Thanks a lot!

Alex

On Wednesday, 2 August 2017 14:44:26 UTC+1, Jonathan Helmus wrote:

[Jonathan Helmus]

The zip issue is a bug in nmrglue that has been fixed in the version on GitHub [1].  You will need to install nmrglue from source if you want to run the example in Python 3.

To accurately determine the FWHM for a peak you will need to fit the peak to some type of lineshape model.  The nmrglue.analysis.lineshapes1d.center_fwhm function will give you a rough estimate of the peak center and FWHM by linear interpolation but this only an approximation, fitting the peak will give you a more accurate estimation.

Cheers,

    - Jonathan Helmu


[1] https://github.com/jjhelmus/nmrglue/pull/73

[Alexandra Grigore]

That's great! Thanks a lot, I had no idea how to "install from source", but I just replaced the nmrglue folder in Python\Python35-32\Lib\site-packages with the one that I downloaded from git and it worked. I will try fitting some lineshapes and calculate the FWHM manually.  

[Alexandra Grigore]

So I figured that the second fitting parameter for a lorenzian lineshape is the FWHM (in points). I managed to get almost the same values as TopSpin gives me with its peakw function. There is a difference when the spectrum is noisy (see attached) because the fitting shifts the baseline a bit higher and not in the middle of the noise. Any idea how I could adjust this?

Thanks a lot, nmrglue is such a life saviour - I will be citing this in the paper I am currently writing.