A performance comparison of modern statistical techniques for molecular descriptor selection and retention prediction in chromatographic QSRR studies

Hancock, Tim; Put, R.; Coomans, Danny; Heyden, Yvan Vander; Everingham, Yvette

doi:10.1016/j.chemolab.2004.11.001

Cited by 131 publications

(60 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Thus, this usually works to decrease undesirable effects of illdistribution of samples. It is reported that, compared with boosting, bagging has a superior advantage to have a robustness against noises [13,31]. On the other hand, boosting is based on the idea of intensive learning of 'bad' samples.…”

Section: Prediction Performances Of the Nir Tobacco Datamentioning

confidence: 98%

“…The main advantage of such ensemble techniques is that they increase the accuracy and stability of the predictor by combining several hypotheses. Such ensemble learning methods have recently been introduced into the field of chemometrics [12,13,15,16]. He et al [12] showed that the boosting tree improves the prediction performance of any single classification method in the classification for chemical data.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Investigations of bagged kernel partial least squares (KPLS) and boosting KPLS with applications to near‐infrared (NIR) spectra

Shinzawa

Jiang

Ritthiruangdej

et al. 2006

Journal of Chemometrics

View full text Add to dashboard Cite

The effects of ensemble learning methods, bagging and boosting, on kernel partial least squares (KPLS) regression are investigated. In bagged KPLS and boosting KPLS, a series of weak hypothesis with picked out samples from a training set are made with KPLS algorithm. Finally, a regression model is developed as an ensemble of these hypotheses. The abilities of bagged KPLS and boosting KPLSR are investigated with two near-infrared (NIR) spectroscopic data sets, NIR diffuse reflectance spectra of dried tobacco leaves and NIR transmission spectra of Thai fish sauces, by comparing with other methods, standard partial least squares (PLS), bagged PLS, and boosting PLS and KPLS. The results reveal that bagged KPLS and boosting KPLS yield superior regression performances to standard PLS. Especially, boosting KPLS indicates clear improvements over to PLS, bagged PLS boosting PLS, KPLS and bagged KPLS.

show abstract

Section: Prediction Performances Of the Nir Tobacco Datamentioning

confidence: 98%

Section: Introductionmentioning

confidence: 99%

Investigations of bagged kernel partial least squares (KPLS) and boosting KPLS with applications to near‐infrared (NIR) spectra

Shinzawa

Jiang

Ritthiruangdej

et al. 2006

Journal of Chemometrics

View full text Add to dashboard Cite

show abstract

“…This randomness has the effect of building new trees with different structures, increasing the variety of relationships modelled within the forest, which in turn improves the overall predictive performance and makes it robust against overfitting. However, while RF are good for identifying main trends, it can lead to some points being poorly predicted (Hancock et al, 2005).…”

Section: Introductionmentioning

confidence: 99%

“…By using many trees, RF have the ability to identify important variables and their relationship with the response. The number of trees, therefore, needs to be set sufficiently high to allow for this convergence as well as to identify the most important variables (Goudarzi and Shahsavani, 2012;Hancock et al, 2005). This randomness has the effect of building new trees with different structures, increasing the variety of relationships modelled within the forest, which in turn improves the overall predictive performance and makes it robust against overfitting.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Advanced analytical tools for the investigation of licit and illicit drug residues in water

Bade¹

View full text Add to dashboard Cite

The University of Queensland, Australia). It is divided into three sections and consists of six scientific articles and one review article (Scientific For my family Article 1).The first section consists of three research articles and deals with the use of high resolution mass spectrometry (HRMS) in the screening and occurrence of PIDs in environmental waters of three countries: Colombia, Italy and Spain. The use of different HRMS instruments (QTOF and LTQOrbitrap) combined with in-house and online databases of up to thousands of compounds enabled numerous compounds to be detected and identified. A continuously-updated in-house database of more than 1000 PIDs and metabolites was made incorporating their exact mass, and also adduct formation, fragment ions and retention time when a standard was available. This entire database was used for the first work, investigating influent and effluent WW as well as SW samples from Bogotá, Colombia, leading to the identification of 40 compounds across all samples (Scientific Article 2). A subset of approximately 100 of the most commonly identified compounds was used in the next work, looking at the use of two HRMS instruments (QTOF and Orbitrap) in the suspect screening of WW and SW samples from Italy and Spain, with up to 28 compounds at least detected (Scientific Article 3). Finally, a non-target screening approach was made using two advanced deconvolution software and incorporating two databases: one of 250 PIDs and the ChemSpider database, with more than 20 compounds being tentatively identified (Scientific Article 4).The second section consists of two research articles and covers the development and utilisation of two retention time prediction approaches in the identification of PIDs in waste and environmental water samples. Initially, a very simple approach was made, using only logKow and a combined in-house and Waters database of 595 reference standards. This led to a prediction accuracy of approximately 70-80% within a retention time window of ±2 minutes. A comparison was made to see whether reducing the mass extraction window incrementally from 50mDa to 5mDa would result in more peaks being removed than the prediction method.Summary ii It was found that even at the narrowest window investigated, this prediction method resulted in almost one quarter of all chromatographic peaks able to be removed as false positives (Scientific Article 5).While this simple approach showed itself quite useful, a more precise method was hoped to aid in the identification of transformation products for which no reference standards are available. Artificial neural networks were thus investigated and found to be much more precise, with up to 95% of all compounds being within a ±2 minute window, leading to the tentative identification of ten metabolites and transformation products and several other compounds for which standards were unavailable (Scientific Article 6).Finally, the third section consists of one research article on the quantitative analysis of new psychoactive substances (NPS) u...

show abstract

Simulating California reservoir operation using the classification and regression‐tree algorithm combined with a shuffled cross‐validation scheme

Yang

Gao

Sorooshian

et al. 2016

Water Resources Research

148

101

View full text Add to dashboard Cite

The controlled outflows from a reservoir or dam are highly dependent on the decisions made by the reservoir operators, instead of a natural hydrological process. Difference exists between the natural upstream inflows to reservoirs and the controlled outflows from reservoirs that supply the downstream users. With the decision maker's awareness of changing climate, reservoir management requires adaptable means to incorporate more information into decision making, such as water delivery requirement, environmental constraints, dry/wet conditions, etc. In this paper, a robust reservoir outflow simulation model is presented, which incorporates one of the well‐developed data‐mining models (Classification and Regression Tree) to predict the complicated human‐controlled reservoir outflows and extract the reservoir operation patterns. A shuffled cross‐validation approach is further implemented to improve CART's predictive performance. An application study of nine major reservoirs in California is carried out. Results produced by the enhanced CART, original CART, and random forest are compared with observation. The statistical measurements show that the enhanced CART and random forest overperform the CART control run in general, and the enhanced CART algorithm gives a better predictive performance over random forest in simulating the peak flows. The results also show that the proposed model is able to consistently and reasonably predict the expert release decisions. Experiments indicate that the release operation in the Oroville Lake is significantly dominated by SWP allocation amount and reservoirs with low elevation are more sensitive to inflow amount than others.

show abstract

A performance comparison of modern statistical techniques for molecular descriptor selection and retention prediction in chromatographic QSRR studies

Cited by 131 publications

References 26 publications

Investigations of bagged kernel partial least squares (KPLS) and boosting KPLS with applications to near‐infrared (NIR) spectra

Investigations of bagged kernel partial least squares (KPLS) and boosting KPLS with applications to near‐infrared (NIR) spectra

Advanced analytical tools for the investigation of licit and illicit drug residues in water

Simulating California reservoir operation using the classification and regression‐tree algorithm combined with a shuffled cross‐validation scheme

Contact Info

Product

Resources

About