Fusion strategies for selecting multiple tuning parameters for multivariate calibration and other penalty based processes: A model updating application for pharmaceutical analysis

Tencate, Alister J.; Kalivas, John H.; White, Abraham

doi:10.1016/j.aca.2016.03.046

Cited by 29 publications

(32 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The processes known as sum of ranking differences (SRD) is used in this study with numerous model quality measures to automatically select calibration tuning parameter values for λ , η , and the number of eigenvectors. The SRD has been shown to be effective in selecting up to 2 calibration tuning parameters and has been well described. Briefly, a matrix of model quality measures is formed with rows designating respective model quality measures and a column for each tuning parameter (tuning parameter triplet in this study).…”

Section: Sample‐wise Calibrationmentioning

confidence: 99%

See 1 more Smart Citation

Sample‐wise spectral multivariate calibration desensitized to new artifacts relative to the calibration data using a residual penalty

Kalivas

Brownfield

Karki

2017

Journal of Chemometrics

Self Cite

View full text Add to dashboard Cite

Calibration maintenance is an important aspect of multivariate calibration. With spectral measurements, the goal of calibration maintenance involves sustaining the predictability of a primary calibration model in new secondary conditions. Among the many methodologies, penalty‐based Tikhonov regularization variants have been successful by sample augmenting primary calibration data with a matrix of just a few secondary samples as well as operating with an additional sparse penalty to include wavelength selection. Studied in this paper is a new sample‐wise (local) Tikhonov regularization–based penalty calibration approach. Penalized is a diagonal matrix with the residual vector (relative to the primary calibration space) of the new secondary sample. Thus, the same full calibration set is used for each new sample. Changing for each secondary sample is the corresponding sample‐wise residual vector on the penalized diagonal matrix. The intent of the presented approach is to form sample‐wise regression vectors desensitized to characteristics of the new sample not present in the primary calibration set. The more distinct the secondary conditions are relative to the primary conditions, the more unsuccessful this local model updating becomes. Proposed is a sample‐wise outlier mechanism to discern when the residual penalty can or cannot be used to form a useful updated model. The residual penalty modeling and outlier detection processes require tuning parameter optimizations. A fusion approach is used to automatically select tuning parameter values. Simulated and near‐infrared data are evaluated, demonstrating the applicability of the method.

show abstract

Section: Sample‐wise Calibrationmentioning

confidence: 99%

“…In previous SRD tuning parameter selection studies, filtering mechanisms were used to eliminate some models from consideration by SRD . Specifically, models were removed from all the possible models across all tuning parameter values not passing specified thresholds.…”

Section: Sample‐wise Calibrationmentioning

confidence: 99%

Sample‐wise spectral multivariate calibration desensitized to new artifacts relative to the calibration data using a residual penalty

Kalivas

Brownfield

Karki

2017

Journal of Chemometrics

Self Cite

View full text Add to dashboard Cite

show abstract

“…Therefore, the smallest SRD value was identified. It should be noted that because of the different scales of the merit value will affect on the ranking results, therefore, the data of each row in the SRD input matrix are normalized with a mean value of 0 and a standard deviation of 1 before SRD computation …”

Section: Theorymentioning

confidence: 99%

“…The sum of ranking differences (SRD) is a recently proposed method to determine the similarity between models and facilitate the selection of models based on their own evaluation merits without considering weight allocation problems . Accordingly, the SRD method was employed to objectively select the optimal parameter combination of the latent variable and the threshold value of the variable importance index in this work.…”

Section: Introductionmentioning

confidence: 99%

Tuning parameter identification for variable selection algorithm using the sum of ranking differences algorithm

Nie

Meng

Chen

et al. 2019

Journal of Chemometrics

View full text Add to dashboard Cite

Variable selection algorithms are often adopted to select the optimal variable from a full set of variables and are efficient for reducing the variable dimension and improving the model accuracy. Nonetheless, the parameters of the variable selection method and regression model, such as the number of latent variables of the partial least squares (PLS) model and the threshold value of the variable importance index, need to be identified. The parameters directly determine the final performance of the model. Currently, these parameters are often determined subjectively. As a result, the model results may be accidental because of the subjective determination of the parameters. To objectively identify these parameters, the sum of ranking differences (SRD) coupled with partial least squares‐variable importance in projection (PLS‐VIP‐SRD) and partial least squares‐uninformative variable elimination (PLS‐UVE‐SRD) algorithms was applied to determine the latent variable of the PLS model and the threshold value of the variable importance index. Furthermore, public near‐infrared data of corn were used as the calculation data. The final results show that the PLS‐VIP‐SRD and PLS‐UVE‐SRD models can more effectively and objectively determine the optimal parameter combination than the PLS‐VIP and PLS‐UVE models. Moreover, the selected variables are easier to interpret, and the prediction accuracy is also improved to some extent.

show abstract

“…It has been applied to various problems so far: from checking the multi-class classification performance in the case of tobacco leaf grades [37] via ranking and classifying chromatographic systems [38,39], to selection of multiple tuning parameters for multivariate calibration [40].…”

Section: Sum Of Ranking Differencesmentioning

confidence: 99%

How to compare separation selectivity of high-performance liquid chromatographic columns properly?

Andrić

Héberger

2017

Journal of Chromatography A

View full text Add to dashboard Cite

Comparison and selection of chromatographic columns is an important part of development as well as validation of analytical methods. Presently there is abundant number of methods for selection of the most similar and orthogonal columns, based on the application of limited number of test compounds as well as quantitative structure retention relationship models (QSRR), from among Snyder's hydrophobicsubtraction model (HSM) have been most extensively used.Chromatographic data of 67 compounds were evaluated using principal component analysis (PCA), hierarchical cluster analysis (HCA), non-parametric ranking methods as sum of ranking differences (SRD) and generalized pairwise correlation method (GPCM), both applied as a consensus driven comparison, and complemented by the comparison with one variable at a time (COVAT) approach. The aim was to compare the ability of the HSM approach and the approach based on primary retention data of test solutes (logk values) to differentiate among ten highly similar C18 columns.The ranking (clustering) pattern of chromatographic columns based on primary retention data and HSM parameters gave different results in all instances. Patterns based on retention coefficients were in accordance with expectations based on columns' physicochemical parameters, while HSM parameters provided a different clustering.Similarity indices calculated from the following dissimilarity measures: SRD, GPCM Fisher's conditional exact probability weighted (CEPW) scores; Euclidian, Manhattan, Chebyshev, and cosine distances Pearson's, Spearman's, and Kendall's, correlation coefficients have been ranked by the consensus based SRD. Analysis of variance confirmed that the HSM model produced statistically significant increases of SRD values for the majority of similarity indices, i.e. HSM transformation of original retention data yields significant loss of information, and finally results in lower performance of HSM methodology. The best similarity measures were obtained using primary retention data, and derived from Kendal's and Spearman's correlation coefficients, as well as GPCM and SRD score values. Selectivity function, Fs, originally proposed by Snyder, demonstrated moderate performance.

show abstract

Fusion strategies for selecting multiple tuning parameters for multivariate calibration and other penalty based processes: A model updating application for pharmaceutical analysis

Cited by 29 publications

References 36 publications

Sample‐wise spectral multivariate calibration desensitized to new artifacts relative to the calibration data using a residual penalty

Sample‐wise spectral multivariate calibration desensitized to new artifacts relative to the calibration data using a residual penalty

Tuning parameter identification for variable selection algorithm using the sum of ranking differences algorithm

How to compare separation selectivity of high-performance liquid chromatographic columns properly?

Contact Info

Product

Resources

About