Estimating the domain of applicability for machine learning QSAR models: a study on aqueous solubility of drug discovery molecules

Schroeter, Timon; Schwaighofer, Anton; Mika, Sebastian; Laak, Antonius ter; Suelzle, Detlev; Ganzer, Ursula; Heinrich, Nikolaus; Müller, Klaus Robert

doi:10.1007/s10822-007-9125-z

Cited by 55 publications

(58 citation statements)

References 32 publications

(40 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Schröter et al, 71 for instance, state that predictions of aqueous solubility for molecules whose structure falls outside the DOA are generally poor. In the literature there are several methods to estimate the DOA of a QSAR model.…”

Section: Resultsmentioning

confidence: 99%

Deep Architectures and Deep Learning in Chemoinformatics: The Prediction of Aqueous Solubility for Drug-Like Molecules

Lusci

Pollastri

Baldi

2013

J. Chem. Inf. Model.

471

447

View full text Add to dashboard Cite

Shallow machine learning methods have been applied to chemoinformatics problems with some success. As more data becomes available and more complex problems are tackled, deep machine learning methods may also become useful. Here we present a brief overview of deep learning methods and show in particular how recursive neural network approaches can be applied to the problem of predicting molecular properties. However molecules are typically described by undirected cyclic graphs, while recursive approaches typically use directed acyclic graphs. Thus we develop methods to address this discrepancy, essentially by considering an ensemble of recursive neural networks associated with all possible vertex-centered acyclic orientations of the molecular graph. One advantage of this approach is that it relies only minimally on the identification of suitable molecular descriptors, since suitable representations are learnt automatically from the data. Several variants of this approach are applied to the problem of predicting aqueous solubility and tested on four benchmark datasets. Experimental results show that the performance of the deep learning methods matches or exceeds the performance of other state-of-the-art methods according to several evaluation metrics and expose the fundamental limitations arising from training sets that are too small or too noisy. A web-based predictor AquaSol is available online through the ChemDB portal (cdb.ics.uci.edu) together with additional material.

show abstract

Section: Resultsmentioning

confidence: 99%

Deep Architectures and Deep Learning in Chemoinformatics: The Prediction of Aqueous Solubility for Drug-Like Molecules

Lusci

Pollastri

Baldi

2013

J. Chem. Inf. Model.

471

447

View full text Add to dashboard Cite

show abstract

“…On the other hand, more sophisticated QSPR models such as those using nonlinear statistical techniques have been proposed that rely solely on calculated molecular descriptors, without the need for the experimentally determined melting point (Palmer et al, 2007;Johnson and Zheng, 2006;Schroeter et al, 2007;Zhou et al, 2008;Huuskonen et al, 1998;Cheng and Merz, 2003;Wassvik et al, 2006;Hou et al, 2004;Bergström, 2005). Melting point is a measure of the crystal lattice energy that needs to be overcome during dissolution, hence the significance in solubility models.…”

Section: Introductionmentioning

confidence: 99%

Estimation of drug solubility in water, PEG 400 and their binary mixtures using the molecular structures of solutes

Ghafourian

Bozorgi

2010

European Journal of Pharmaceutical Sciences

View full text Add to dashboard Cite

With the aim of solubility estimation in water, polyethylene glycol 400 (PEG) and their binary mixtures, Quantitative Structure -Property Relationships (QSPRs) were investigated to relate the solubility of a large number of compounds to the descriptors of the molecular structures. The relationships were quantified using linear regression analysis (with descriptors selected by stepwise regression) and Formal Inference -based Recursive Modeling (FIRM). The models were compared in terms of the solubility prediction accuracy for the validation set. The resulting regression and FIRM models employed a diverse set of molecular descriptors explaining crystal lattice energy, molecular size, and solute -solvent interactions. Significance of molecular shape in compound's solubility was evident from several shape descriptors being selected by FIRM and stepwise regression analysis. Some of these influential structural features, e.g. connectivity indexes and Balaban topological index, were found to be related to the crystal lattice energy. The results showed that regression models outperformed most FIRM models and produced higher prediction accuracy. However, the most accurate estimation was achieved by the use of a combination of FIRM and regression models. The results also showed that the use of melting point in regression models improves the estimation accuracy especially for solubility in higher concentrations of PEG. Aqueous or PEG/water solubilities can be estimated by these models with root mean square error of below 0.70.

show abstract

“…evaluate several distance‐based measures to estimate the domain of applicability of QSAR models, as well as more sophisticated approaches. The applicability of Gaussian process models can be estimated using their built‐in estimate of predictive variance 23…”

Section: Related Workmentioning

confidence: 99%

Visual Interpretation of Kernel‐Based Prediction Models

Hansen

Baehrens

Schroeter³

et al. 2011

Molecular Informatics

Self Cite

View full text Add to dashboard Cite

Statistical models are frequently used to estimate molecular properties, e.g., to establish quantitative structure-activity and structure-property relationships. For such models, interpretability, knowledge of the domain of applicability, and an estimate of confidence in the predictions are essential. We develop and validate a method for the interpretation of kernel-based prediction models. As a consequence of interpretability, the method helps to assess the domain of applicability of a model, to judge the reliability of a prediction, and to determine relevant molecular features. Increased interpretability also facilitates the acceptance of such models. Our method is based on visualization: For each prediction, the most contributing training samples are computed and visualized. We quantitatively show the effectiveness of our approach by conducting a questionnaire study with 71 participants, resulting in significant improvements of the participants' ability to distinguish between correct and incorrect predictions of a Gaussian process model for Ames mutagenicity.

show abstract

Estimating the domain of applicability for machine learning QSAR models: a study on aqueous solubility of drug discovery molecules

Cited by 55 publications

References 32 publications

Deep Architectures and Deep Learning in Chemoinformatics: The Prediction of Aqueous Solubility for Drug-Like Molecules

Deep Architectures and Deep Learning in Chemoinformatics: The Prediction of Aqueous Solubility for Drug-Like Molecules

Estimation of drug solubility in water, PEG 400 and their binary mixtures using the molecular structures of solutes

Visual Interpretation of Kernel‐Based Prediction Models

Contact Info

Product

Resources

About