The PLS Approach

Using our dataset which includes all of the descriptors mentioned so far, we conducted a PLS analysis using SIMCA software [34], In the initial PLS model, MW, V, and a (Alpha) were removed because they are in each case highly correlated with CMR (r 0.95). SIMCA s VIP function selected only qmin (Qnegmin) for removal on the basis of it making no important contribution to the model. In the second model, 2q+/a (SQpos A) and ECa/a (SCa A) coincided nearly exactly in the three-component space of these two, we decided to keep only ECa/a in the third and final model. This model consisted of three components and accounted for 75% of the variance in log SQ the Q2 value was 0.66. 

We report here studies conducted with 58 crystalline drug-like compounds. The solubilities are of high quality, having been determined in only three laboratories using the same automated potentiometric method. However, the set is too small to 

We thank Professor Per Artursson of Uppsala University for allowing us to use the intrinsic solubilities reported in reference [28] prior to their publication, and Suzanne Tilton and Alexis Foreman of pION INC. for some of the pSOL solubility measurements. 

1 McFariand, J. W., Avdeef, A., Berger, C. M., Raevsky, O. A., Estimating the water solubilities of crystalline compounds from their chemical structures alone, J. Chem. Inf. Comput. Sci. 2001, 41, 1355— 1359. 

---

The PLS approach to multivariate linear regression modeling is relatively new and not yet fully investigated from a theoretical point of view. The results with calibrating complex samples in food analysis 122,123) j y jnfj-ared reflectance spectroscopy, suggest that PLS could solve the general calibration problem in analytical chemistry. 

PLS was frequently used to connect the variation in chemical stiucture of polypeptides to variation of their biological activities. Chemical stiucture of peptides were characterized by Zi, Z2 and Z3 deseriptors [45, 189, 209, 210] of their amino acid moieties. The PLS approach gave significant cross-vahdated predictions of the activities in most examples [191, 188, 189]. Sjostrom et al. [193] analysed the information content of signal peptide amino acid sequences fromproteins. Other applications are available in the hterature [45,61-64,192,222, 225-228]. The combined use of PLS and pattern recognition methods [229, 230] is reviewed in [183,194]. 

A method that accounts for the concentration-spectra relationships during decomposition is the PLS approach. 

On the face of it, PLS appears to offer a much superior approach to the construction of linear regression models than MLR or PCR (since the dependent variable is used to construct the latent variables) and for some data sets this is certainly true. Application of PLS to the charge-transfer data set described in the last section resulted in a PLS model containing only two dimensions which explained over 90 per cent of the variance in the substituent constant data. This compares very favourably with the two- and three-dimensional PCR equations (eqns 7.7 and 7.8) which explain 73 and 81 per cent of the variance respectively. Another advantage that is claimed for the PLS approach is its ability to handle redundant information in the independent variables. Since the latent variables are constructed so as to correlate with the dependent variable, redundancy in the form of colli-nearity and multicollinearity in the descriptor set should not interfere. This is demonstrated by fitting PLS models to the 31 variable and 11 variable parameter sets for the charge-transfer data. As shown in Table 7.7 the resulting PLS models account for very similar amounts of variance in k. 

As shown by Kvalheim (6), approaches to decomposition of multivariate data in terms of latent variables can be developed within the frame of a generalized NIPAL algorithm (7), for instance, decomposition into principal components (8) and decomposition using the PLS approach (9,10). 

The following will be established. At high temperatures, we have microstates in which each component is finite in size hence, the multiplicity of each bond is even. This condition then uniquely defines disordered microstates. There are two distinct percolating components = N). One ofthem has all bond multiplicities even. We will treat this to represent a disordered microstate for the obvious reason. The other percolating component has all bond multiplicities odd. This is obviously a component with a different symmetry, and we will see below that it represents the ordered state. The possible odd multiplicities of its bonds uniquely determine ordered microstates. The sets of disordered and ordered microstates are obviously disjoint, which ensures that there will be no stable nuclei in the metastable state if it occurs, as discussed in Section 10.1.6, and distinguishes our approach from the PL approach. Because of distinct symmetries of the two phases, our model genuinely represents a melting transition, and not a liquid-gas transition in which the symmetry remains the same in both equilibrium phases. 

The PLS approach was developed around 1975 by Herman Wold and co-workers for the modeling of complicated data sets in terms of chains of matrices (blocks), so-called path models . Herman Wold developed a simple but efficient way to estimate the parameters in these models called NIPALS (nonlinear iterative partial least squares). This led, in turn, to the acronym PLS for these models, where PLS stood for partial least squares . This term describes the central part of the estimation, namely that each model parameter is iteratively estimated as the slope of a simple bivariate regression (least squares) between a matrix column or row as the y variable, and another parameter vector as the x variable. So, for instance, in each iteration the PLS weights w are re-estimated as u X/(u u). Here denotes u transpose, i.e., the transpose of the current u vector. The partial in PLS indicates that this is a partial regression, since the second parameter vector (u in the... 

The PLS approach, where the objectives of the analysis are quantified in a matrix Y, which is then used to guide the... 

B being a matrix of regression coefficients, the PLS approach (see Chapter 4) can be used to calculate the model even in the cases where the matrix X is ill-conditioned (presence of collinearity or low samples/variables ratio). The corresponding classification method is then called partial least squares-discriminant analysis (PLS-DA). 

Some methods that paitly cope with the above mentioned problem have been proposed in the literature. The subject has been treated in areas like Cheraometrics, Econometrics etc, giving rise for example to the methods Partial Least Squares, PLS, Ridge Regression, RR, and Principal Component Regression, PCR [2]. In this work we have chosen to illustrate the multivariable approach using PCR as our regression tool, mainly because it has a relatively easy interpretation. The basic idea of PCR is described below. 

The first theoretical handling of the weak R-T combined with the spin-orbit coupling was carried out by Pople [71]. It represents a generalization of the perturbative approaches by Renner and PL-H. The basis functions are assumed as products of (42) with the eigenfunctions of the spin operator conesponding to values E = 1/2. The spin-orbit contribution to the model Hamiltonian was taken in the phenomenological form (16). It was assumed that both interactions are small compared to the bending vibrational frequency and that both the... 

S Wold, A Ruhe, H Wold, WJ Dunn III. The collmearity problem m linear regression. The partial least squares (PLS) approach to generalized inverses. SIAM I Sci Stat Comput 5 735-743, 1984. 

In recent years there has been much activity to devise methods for multivariate calibration that take non-linearities into account. Artificial neural networks (Chapter 44) are well suited for modelling non-linear behaviour and they have been applied with success in the field of multivariate calibration [47,48]. A drawback of neural net models is that interpretation and visualization of the model is difficult. Several non-linear variants of PCR and PLS regression have been proposed. Conceptually, the simplest approach towards introducing non-linearity in the regression model is to augment the set of predictor variables (jt, X2, ) with their respective squared terms (xf,. ..) and, optionally, their possible cross-product... 

All of the above would seem to indicate that I am totally against using MLR. This is not the case. In my practice, I always try the simplest approach first. This means first trying MLR. If that does not work, then I use PLS. If that does not work -well, some people may use neural networks, but I have not yet found a need to do so. I think you are right in saying that there has been a lot of hype over PLS (although not as much as there has been over neural nets ) In many cases MLR works great, and I will continue to use it. To paraphrase Einstein, Always use the simplest approach that works - but no simpler. ... 

The point being that, as our conclusions indicate, this is one case where the use of latent variables is not the best approach. The fact remains that with data such as this, one wavelength can model the constituent concentration exactly, with zero error - precisely because it can avoid the regions of nonlinearity, which the PCA/PLS methods cannot do. It is not possible to model the constituent better than that, and even if PLS could model it just as well (a point we are not yet convinced of since it has not yet been tried -it should work for a polynomial nonlinearity but this nonlinearity is logarithmic) with one or even two factors, you still wind up with a more complicated model, something that there is no benefit to. 

The truly parallel approach employed in the design of the /./Pl.C system described in this chapter produces a clear reduction in analysis time when compared with traditional HPLC techniques. For example, a separation method of 5 min duration in a //Pl.C system would allow simultaneous evaluation of 24 samples within that time—an average analysis time of 12.5 sec/sample. Similarly, the dimensions of the columns housed in the cartridge require smaller amounts amount of solvent (mobile phases) for the analysis. 

Although all obsidian samples exhibited quite similar broadband LIBS spectra, the PLS-DA approach was able to clearly discriminate samples from the 5 different California obsidian locations. Within CVF, it is possible to distinguish the five subgroupings defined on the basis of ICP-MS... 

The above approaches used the idea of conjugation length control in PTs by distorting the polymer backbone with bulky substituents as side groups. Hadziioannou and coworkers [509,510] demonstrated PL and EL tuning via exciton confinement with block copolymers... 

---