Generative Topographic Mapping (GTM): Universal Tool for Data Visualization, Structure‐Activity Modeling and Dataset Comparison

Abstract

Here, the utility of Generative Topographic Maps (GTM) for data visualization, structure-activity modeling and database comparison is evaluated, on hand of subsets of the Database of Useful Decoys (DUD). Unlike other popular dimensionality reduction approaches like Principal Component Analysis, Sammon Mapping or Self-Organizing Maps, the great advantage of GTMs is providing data probability distribution functions (PDF), both in the high-dimensional space defined by molecular descriptors and in 2D latent space. PDFs for the molecules of different activity classes were successfully used to build classification models in the framework of the Bayesian approach. Because PDFs are represented by a mixture of Gaussian functions, the Bhattacharyya kernel has been proposed as a measure of the overlap of datasets, which leads to an elegant method of global comparison of chemical libraries.

Generative Topographic Mapping of Conformational Space

“…Herein, the in-house ISIDA GTM implementation was used. [17] In GTM, a point in the low-dimensional (usually 2D) latent space (LS) corresponds to an image on the manifold embedded in the initial CS. The manifold is defined by a mapping function y(x; W) assessed with the help of M radial basis functions (RBFs) of width w regularly distributed in LS.…”

“…Setup and performances of top maps with best property prediction propensities. Reported key setup parameters: are conformational descriptors used, #Nodes (per grid line; GTM grid being a square of #Nodes 3 #Nodes), #RBF (number of Radial Basis Functions), Regularization coefficient l and Gaussian width parameter w (please refer to cited publications [17] for the specific meaning of the latter). The right-hand columns report root-mean-squared errors RMSE and below, in parentheses, cross-validated determination coefficients Q 2 for the 5 properties which the maps were challenged to model: total, non-bonded, contact energies (in kcal/mol), solvent-accessible surface area ( 2 ) and contact fingerprint darkness (percentage of fulfilled contacts out of the total number of putative contacts monitored by CONTFP).…”

Application of the methods of molecular modeling to the search for new biologically active substances

“…Результати, одержані на попередніх етапах дослідження (сет активних і неактивних речо-вин, результати віртуального скринінгу), були застосовані для побудови моделей GTM згідно з алгоритмом, запропонованим у роботах [30,31]. Оцінювали такі параметри: FreE (загальна вільна енергія комплексоутворення), Cntc (кон-тактна енергія зв'язування між усіма молекула-ми ліганду і протеїну), Hbnd (енергія Н-зв'язків між протеїном і лігандом), Bump (енергія сте-ричних зіткнень), Intl (енергія напруженості ліганду).…”

Visualization of multi-property landscapes for compound selection and optimization

“…Rather, different visualization techniques have also been introduced to generalize chemical space display including, for example, similarity-based compound networks [12] and molecular layout algorithms [13] for smaller data sets, projections from high-dimensional descriptors spaces based on principal component analysis for large (or very large) data sets [14,15], and generative topographic mapping (GTM) [16]. GTM was designed to project from high-dimensional feature spaces onto latent 2D space representations in which points (nodes) correspond to normal probability distributions derived from the original data space that determine the mapping of compounds to the latent space.…”