|Roma Tauler||Federico Marini|
Course on Multivariate Curve Resolution (MCR)
Professor Romà Tauler
IDAEA-CSIC, Jordi Girona 18-26, 08034 Barcelona, Spain
Basic knowledge of chemometrics(multivariateanalysis of chemical data)
This course is dedicated to presenting the theoretical aspects of the MCR method and to its practical applications. Procedures are described for exploring the structure of the data, generation of initial estimates, choosing appropriate constraints, and to the application of MCR to data sets arranged in a table or matrix or to data multisets arranged in multiple data tables, data matrices, data cubes or data hypercubes with multiple directions or ways (three-way, multi-way data). Theoretical aspects will be accompanied with various selected examples. Ways to handle and ascertain the presence of uncertainties in MCR results due to rotation ambiguities or to noise propagation will be described. Extensions of the MCR-ALS method to the analysis of kinetic reaction data will be also introduced.
- The MCR bilinear model. The MCR-ALS method. Number of components, initial estimates and constraints (non-negativity, unimodality, mass balance, selectivity / local rank, etc).
- MCR application to the study of a) chemical reactions and processes, b) hyphenated chromatography, c) hyperspectral imaging, d) environmental data; e) metabolomics/lipìdomics data
- MCR for multiset and multiway data analysis (multiset and multiway data). Multilinear models and implementation of constraints (trilinear, quadrilineal, interaction …). MCR for multidimensional spectroscopy and chromatography.
- MCR Quantitative. MCR ambiguity: MCR-BANDS. Handling data uncertainties: MLPCA-MCR-ALS. MCR hard and soft modelling analysis of kinetic data
Course on Predictive Chemometric Modeling
Professor Federico Marini
Dept. of Chemistry, University of Rome La Sapienza, P.le Aldo Moro 5, I-00185 Rome, Italy
Very basic knowledge of chemometrics(at least principal component analysis) is recommended but not mandatory
This course aims at introducing the chemometric tools for building predictive models, i.e., models which relate a (usually measured) profile to one or more responses one is interested in estimating. Depending on whether the nature of the response to be predicted is qualitative (it can take only a limited set of discrete values) or quantitative (real-valued), one usually speaks of classification or calibration/regression. In this respect, the course will introduce the main theoretical bases of predictive modeling and discuss the chemometric tools more frequently used in the framework of calibration and classification, focusing in particular on those methods which are based on projecting the data onto a reduced space of latent variables (bilinear models). Particular attention will also be devoted to the proper validation of the models, both in terms of predictive figures of merit and interpretation of the results. All the main concepts and methods discussed will be illustrated by means of practical examples.
- General introduction on predictive modeling. The concept of calibration.
- Least squares methods for univariate and multivariate data. Evaluating the regression models in terms of predictive ability. Interpreting the regression coefficients.
- The need for bilinear regression: Principal component regression (PCR) and Partial least squares regression (PLS).
- Predicting qualitative attributes: introduction to classification. Discriminant vs modeling approaches.
- Partial least squares discriminant analysis (PLS-DA) for discriminant classification
- SIMCA and class-modeling
- Validation: an essential step.