Adv. Analytical Chemistry Lectrure 24 - MVA.ppt

* * * * * * Does Our Training Set Accurately Describe Our Unknown Data? -0.4 V -0.4 V 1.3 V 15 s DA Info pH Info None of the Above -0.4 V -0.4 V 1.3 V 15 s 15 s 15 s Qα Threshold for Significance Jackson, J.E.; Mudholkar, G.S. Technometrics, 1979, 21, 341-349. None of the Above -0.4 V -0.4 V 1.3 V 15 s Qα represents a threshold where 95% of noise would fall under. It is a threshold for significance, NOT confidence! If Q crosses Qα, the training set is insufficient to analyze the measured data. DA pH Find Relevant PCs Discard Noise PCs Calibrate PCs Using Training Set Data Project onto PCs to Predict Concentrations Principal Component Regression Plot Training Set in an Intensity-Based System 250 nM DA 0.05 pH pH Residual 15 s -0.4 V -0.4 V 1.3 V 4.0 -2.7 Current (nA) (Qα) Multivariate Curve Resolution-Alternating Least Squares (MCR-ALS) D Spectra Conc(t) De Juan, A.; Tauler, R. Crit. Rev. Anal. Chem., 2006, 36, 163-176. Original profile Original profile Vertical Horizontal Data Structure Constraints Non-negativity Unimodality Mass spectrum before selectivity is applied Mass spectrum after selectivity is applied Multivariate Curve Resolution-Alternating Least Squares (MCR-ALS) D = C ? ST + E C = D ? S ? (S ? ST)-1 ST = (CT ? C)-1 ? CT ? D R. Tauler, Chemom. Intell. Lab. Systems, 30, 133-146, 1995. D = data C = concentration change per time S = spectra E = error 1. Give an initial guess for spectral information (S) to solve for concentration changes (C). Multivariate Curve Resolution-Alternating Least Squares (MCR-ALS) D = C ? ST + E C = D ? S ? (S ? ST)-1 ST = (CT ? C)-1 ? CT ? D R. Tauler, Chemom. Intell. Lab. Systems, 30, 133-146, 1995. D = data C = concentration change per time S = spectra E = error 2. Constrain C and use the result to calculate a new S. D = C ? ST + E C = D ? S ? (S ? ST)-1 ST = (CT ? C)-1 ? CT ? D R. Tauler, Chemom. Intell. Lab. Systems, 30, 133-146, 1995. D = data C = concentration change per time S = spectra E = error Multivar