| Author | Message | ||
     Fernando Schimidt (dash_9) New member Username: dash_9 Post Number: 1 Registered: 6-2012 |
Hello everyone, I found these articles on the subject: 1) Pure Appl. Chem., Vol. 78, No. 3, pp. 633–661, 2006 2)Spectroscopy 18(12) December 2003, pp. 11-114 3)Analytica Chimica Acta, 423, 2000, pp. 41–49. Best regards Fernando | ||
| David Didion (david_umons) New member Username: david_umons Post Number: 4 Registered: 12-2011 |
Dear all, I allow myself to come back on this topic since no answer has been given to Andre's question. Thank you in advance. David | ||
| Andre RIEUTORD |
Dear colleagues, What formula would you use to derive LOD or LOQ from SEP (or CSet SEE, SECV) computed with PLS ? | ||
| Bruce H. Campbell (Campclan) |
Andre, I wouldn't use any of them. In my opinion, the best way to determine both the limit of detection and limit of quantitation is from th application of the calibration to samples. The pattern to be used varies in the examples I have seen, but I try to use at least six samples and at least three determinations each. I have found that the standard deviations calculated from such a procedure do not vary much on a day-to-day set of measurements. Bruce | ||
| hlmark |
There's a very active disucssion going on in the Chemometrics community about just this topic: what is the meaning and how do you define LOD for multivariate analytical methods? Reducing it to a univariate situation is OK in some cases, but not in all. Basically, changing the situation to the univariate one by doing something akin to what Bruce suggests amounts to determining the LOD over all possible types of sample. In classical analysis, this is what is called "matrix effect" (where the matrix in this case is the chemical matrix that the analyte exists in, as distinguished from a mathematical matrix that defines computations). In classical univariate analysis, the "matrix effect" is simply considered to be one more source of error. In contrast, multivariate analysis allows (in theory, at least) compensation for the matrix effect. The benefit is that in favorable cases you could attain a lower LOD, compared to a univariate analysis. The downside is that the LOD itself then becomes sensitive to the nature of the matrix, with different matrices giving different values of LOD for the analyte. At the very least, then, actually achieving the benefits of multivariate analysis indicates that you must have two models: one for the concentration of the analyte and a separate one for the LOD. In this view, the reduction to univariate analysis is essentially equivalent to finding some sort of "average" value of the LODs over all different possible matrices. Some of the mathematically-oriented chemometricians are working all this out. Howard \o/ /_\ |