I don't use OPUS. Can you upload before and after images of your deconvolution? Just guessing here, but is the intensity equivalent to gain, and is width equivalent to a point spread function (PSF)? Also, does OPUS give you a choice of deconvolution algorithms which you can use. Most of the originally developed image deconvolution algorithms (such as Lucy-Richardson) result in fourier "ringing" around high contrast points in the deconvolved image such that values can easily become negative. More recently developed deconvolution algorithms attempt to solve this problem.

Dear Mangesh,

Please look at some of our publications on ovalbumin, BSA and Glycinin at different concnetration

Several article is out there to give you insight how to do it. Your initial consideration looks correct. Read some of my article on curve fitting of biomolecules: Quality Assessment of Recombinant Proteins by Infrared Spectroscopy. Characterisation of a Protein Aggregation Related Band of the Ca2+-ATPase. Analyst, 2014.

Characterization of recombinant antibodies for cancer therapy by infrared spectroscopy. Biologicals, 2013, 1-7

Dear Mangesh

Try first to make the curve fitting without deconvolution based  on the second derivative. When doing the curve fitting with Omnic software, for example, the half band width is chosen according to the narrowest band in the second derivative.

Dear All,

Thanks for your valuable comments. I think, baseline correction from 1600-1700 cm-1, followed secondary derivative, peaks obtained in the secondary derivatives will be used for the curve fitting, value of half band width will be put in the width column followed by curve fit iterations, this procedure for the secondary structure quantification i have got after going through all of urs comments. My RMS value is coming in the range of 0.000093 by using this procedure.  Please comment if anything is not upto the mark in this procedure

Hi Mangesh,

I'm also an Opus user - but whatever software you use, please don't accept it as a "black box", make sure you really understand what it's doing. If necessary, use a mathematics software like Matlab or Igor Pro to check that the calculation procedure is according to your wishes. Two points in particular:

1. How many bands should you use to fit? It's not correct to use the quality of the fit (i.e. residual RMS error) to determine this, because the fit quality will always increase with increasing number of bands ad infinitum, regardless of whether there's any connection with the actual number of bands present in your data - this is simply the difference between mathematics and real life.

So you need an empirical determination of how many peaks your data really justifies. Two alternative ways to do this:

a) Take the second derivative of your data and count the minima. See the 3 attached JPGs, which illustrate how this works for a single synthetic Gauss peak and for two peaks with and without overlap. Note that when the two peaks are near to each other, the 2nd derivative shows the smaller peak shifted from its true position - but probably still good enough as an initial estimate for the iterative fitting procedure. This problem disappears when the peaks are far apart, but a spurious additional minimum appears in between due to the interaction of the side bands. In spite of these potential issues, at least you're using an empirical procedure to count the bands which others can reproduce.

b) Fit with 1 peak, then 2 peaks, then 3 peaks, etc., watching how the residual RMS error decreases with each step. Decide in advance what you consider to be a significant improvement, e.g. 5%. Stop adding peaks when the improvement in the RMS error value falls below this threshold. What you're saying here is that you're describing your data with this number of peaks because the data does not contain significant evidence for any more peaks than that - as a scientist I find that intellectually satisfying.

2. The second important issue is to be careful of what you're doing with the baseline subtraction and spectral region cutoff. Are you sure that it's really some kind of baseline artefact underlying your amide I data, and not the flanks of peaks outside that cutoff range, especially at the 1600/cm end? Look again at my "2 peaks near.jpg" example and imagine subtracting a straight "baseline" from -2 to +2, then fitting just that region. The remainder after subtraction would no longer be Gaussian curves and the fit will doing something arbitrary and wrong to try explaining the shape of the curve. This may well explain why you're getting negative peaks.

The solution? Ideally, you do the experiment carefully and properly so that the data has no baseline artefact. If that's not possible, you understand the cause of the artifact - scattering? residual water? - and describe it with the correct equation. Assume the peaks towards the edges of your selected spectral region will always come out wrong because of the cutoff, therefore fit from 1500 to 1800/cm and then discard the peaks outside the region of interest.    

Cheers,

David

Dear Mangesh,

do not only restrict/rely on the curve fitting/deconvolution concept to extract information from (ATR) FTIR spectra of proteins in the the Amide I band region.

Also use Partial Least Square (PLS) based approaches also known since 1990.

For example Pezolet, Chapman, Rahmelow [1, 2, 3] treated FTIR protein spectral data by Factor Analysis, which is based on the representation of the Amide I band of a protein with unknown structure (A) by a linear combination of several factor spectra of proteins with known structure (B, C, D):

A_unknown = b * B_known + c * C_known + d * D_known

In principle least square and other numerical methods are used to iteratively optimize the factors b, c and d,  like it is common for the analysis of circular dichroism spectra on proteins.From b, c, and d you can calculate the secondary structure portions (alpha, beta, random etc.).

1. F. Dousseau, M. Pezolet, Biochemistry 1990, 29, 8771-8779.

2. D.C. Lee, P.I. Haris, D. Chapman, R.C. Mitchell, Biochemistry 1990, 29, 9185-9193.

3. K. Rahmelow, W. Hübner, Anal. Biochem. 1996, 241(1), 5-13.

Best regards,

Martin

How to do this by origin software?