Comparison of Different Approaches to Define the Applicability Domain of QSAR Models

Sahigara, F; Mansouri, Kamel; Ballabio, Davide; Mauri, Andrea; Consonni, Viviana; Todeschini, Roberto

doi:10.3390/molecules17054791

Cited by 457 publications

(403 citation statements)

References 23 publications

Supporting

Mentioning

396

Contrasting

Unclassified

Order By: Relevance

“…This may be attributed to the alignment of a number of factors, including increased availability of data, advances in data-mining methodologies as well as a more widespread appreciation of how to avoid many of the numerous pitfalls in building and applying QSAR models (Cherkasov et al 2014). Current trends in the field include efforts in chemical data curation (Williams et al 2012), automation of QSAR model building (Cox et al 2013), exploration of alternative descriptors (Cherkasov et al 2014), and efforts to help define the Applicability Domain (AD) of a given QSAR model (Sahigara et al 2012).…”

Section: Discussionmentioning

confidence: 99%

Meta-QSAR: a large-scale application of meta-learning to drug design and discovery

et al. 2017

View full text Add to dashboard Cite

We investigate the learning of quantitative structure activity relationships (QSARs) as a case-study of meta-learning. This application area is of the highest societal importance, as it is a key step in the development of new medicines. The standard QSAR learning problem Learn (2018) 107:285-311 is: given a target (usually a protein) and a set of chemical compounds (small molecules) with associated bioactivities (e.g. inhibition of the target), learn a predictive mapping from molecular representation to activity. Although almost every type of machine learning method has been applied to QSAR learning there is no agreed single best way of learning QSARs, and therefore the problem area is well-suited to meta-learning. We first carried out the most comprehensive ever comparison of machine learning methods for QSAR learning: 18 regression methods, 3 molecular representations, applied to more than 2700 QSAR problems. (These results have been made publicly available on OpenML and represent a valuable resource for testing novel meta-learning methods.) We then investigated the utility of algorithm selection for QSAR problems. We found that this meta-learning approach outperformed the best individual QSAR learning method (random forests using a molecular fingerprint representation) by up to 13%, on average. We conclude that meta-learning outperforms base-learning methods for QSAR learning, and as this investigation is one of the most extensive ever comparisons of base and meta-learning methods ever made, it provides evidence for the general effectiveness of meta-learning over base-learning.

show abstract

Section: Discussionmentioning

confidence: 99%

Meta-QSAR: a large-scale application of meta-learning to drug design and discovery

et al. 2017

View full text Add to dashboard Cite

show abstract

“…geometry, range, distance or probability density function based approaches) proposing to define the applicability domain region of statistical models based on different algorithms. For more detailed information about the available approaches for defining the (Q)SAR model applicability domain, interested readers are encouraged to refer to the review papers by others (Jaworska, Aldenberg, & Nikolova, 2005;Sahigara, et al, 2012).…”

Section: Support Vector Machines (Svm)mentioning

confidence: 99%

(Q)SAR modelling of nanomaterial toxicity: A critical review

et al. 2015

View full text Add to dashboard Cite

There is an increasing recognition that nanomaterials pose a risk to human health, and that the novel engineered nanomaterials (ENMs) in the nanotechnology industry and their increasing industrial usage poses the most immediate problem for hazard assessment, as many of them remain untested. The large number of materials and their variants (different sizes and coatings for instance) that require testing and ethical pressure towards non-animal testing means that expensive animal bioassay is precluded, and the use of (quantitative) structure activity relationships ((Q)SAR) models as an alternative source of hazard information should be explored.(Q)SAR modelling can be applied to fill the critical knowledge gaps by making the best use of existing data, prioritize physicochemical parameters driving toxicity, and provide practical solutions to the risk assessment problems caused by the diversity of ENMs. This paper covers the core components required for successful application of (Q)SAR technologies to ENMs toxicity prediction, and summarizes the published nano-(Q)SAR studies and outlines the challenges ahead for nano-(Q)SAR modelling. It provides a critical review of (1) the present status of the availability of ENMs characterization/toxicity data, (2) the characterization of nanostructures that meets the need of (Q)SAR analysis, (3) the summary of published nano-(Q)SAR studies and their limitations, (4) the in silico tools for (Q)SAR screening of nanotoxicity and (5) the prospective directions for the development of nano-(Q)SAR models.

show abstract

“…To obtain predictions for any incoming sample set using the model, the first group of methods are used to ensure that the compounds are within the so-called 'active subspace': which essentially means that we are actually performing interpolation only, not extrapolation. [52,53] For the distancebased approach, a pre-defined statistic is calculated to quantify the degree of nearness of the test compounds to the training set and based on whether that statistic is above or below a certain cutoff value, predictions for those compounds are considered acceptable or not. [52,54] …”

Section: Applicability Domain Of Qsar Modelsmentioning

confidence: 99%

“…[52,53] For the distancebased approach, a pre-defined statistic is calculated to quantify the degree of nearness of the test compounds to the training set and based on whether that statistic is above or below a certain cutoff value, predictions for those compounds are considered acceptable or not. [52,54] …”

Section: Applicability Domain Of Qsar Modelsmentioning

confidence: 99%

Use of Graph Invariants in Quantitative Structure-Activity Relationship Studies

Basak¹

2016

Croat. Chem. Acta

View full text Add to dashboard Cite

This chapter reviews results of research carried out by Basak and collaborators during the past four decades or so in the development of novel mathematical chemodescriptors and their applications in quantitative structure-activity relationship (QSAR) studies related to the prediction of toxicities and bioactivities of chemicals. For chemodescriptors based QSAR studies, we have used graph theoretical, three dimensional (3-D), and quantum chemical indices. The graph theoretic chemodescriptors fall into two major categories: (a) Numerical invariants defined on simple molecular graphs representing only the adjacency and distance relationship of atoms and bonds; such invariants are called topostructural (TS) indices; (b) Topological indices derived from weighted molecular graphs, called topochemical (TC) indices. Collectively, the TS and TC descriptors are known as topological indices (TIs). The set of independent variables used for modeling also includes a group of threedimensional (3-D) molecular descriptors. Semi-empirical and various levels of ab initio quantum chemical indices have also been used for hierarchical QSAR (HiQSAR) modeling. Results indicate that in many cases of property / activity / toxicity analyzed by us, a TS + TC combination explains most of the variance in the data.

show abstract

Comparison of Different Approaches to Define the Applicability Domain of QSAR Models

Cited by 457 publications

References 23 publications

Meta-QSAR: a large-scale application of meta-learning to drug design and discovery

Meta-QSAR: a large-scale application of meta-learning to drug design and discovery

(Q)SAR modelling of nanomaterial toxicity: A critical review

Use of Graph Invariants in Quantitative Structure-Activity Relationship Studies

Contact Info

Product

Resources

About