Non-linear peak deconvolution

Short description:

In chromatographic analysis, when peaks are partially overlapped, we have several methods to separate mathematically what couldn’t be completely separated physically. A family of methods arises when we consider hyphenated techniques (e.g., HPLC-DAD or GC-MS). Almost all of these techniques rely on the bilinear nature of the data. However, when the detection is single-channeled (i.e. no hyphenation present), the methods to separate the signal of the different co-eluting peaks are of different nature. The majority of those methods involve accommodating the peak signal to some mathematical model (e.g., Gaussian, Exponentially-modified Gaussian, etc.). Normally these methods suffer from lack-of-fit, since unfortunately all peaks in chromatography are are not exactly accommodating to the proposed models.

Things can get a little bit better if data from several samples is considered all-together. The method benefits from some relational aspects (like for example the peak shape, which is normally constant through injections if the chromatographic system is sufficiently stable), averaging out (random) deviations from the proposed model. A first aspect of this project was to develop the generic method for peak deconvolution using data from several injections [6a]. Given the ammount of parameters to optmisize and the possibilities of existence of local optima (remember that the peak fitting is non-linear with respect to the parameters), a genetic algorithm was used [2a]. In a second phase we developed an automated method for peak deconvolution, which in fact consisted on two parts: peak detection [14a] and peak deconvolution [13a]. A side project was to develop a resolution metric that was able to evaluate the potential capabilities of the application of deconvolution [9a]. The idea was to use this measurement as objective function in chromatographic optimisation, so the user is directed towards those situations in which deconvolving the peaks has the greatest possibility of success. It turned out that a measurement based on the net-analyte signal was the best method to assess this “deconvolution-oriented” resolution metric. to The method was applied to several application areas involving HPLC.

The peak detection method developed [14a] (based on derivatives) has been later on applied as a first step for peak detection in two-dimensional chromatography.


This project was developed at the University of Valencia and at the Free University of Brussels. Several people were involved. See authors of the publications for more details about authorship.


University of ValenciaFree University of Brussels.


See my presentation in Brussels (ChemoAC-2004) for the aspects concerning automation of the method.


Not available.