One‐ to Four‐Dimensional Kernels for Virtual Screening and the Prediction of Physical, Chemical, and Biological Properties.

Abstract

Many chemoinformatics applications, including high-throughput virtual screening, benefit from being able to rapidly predict the physical, chemical, and biological properties of small molecules to screen large repositories and identify suitable candidates. When training sets are available, machine learning methods provide an effective alternative to ab initio methods for these predictions. Here, we leverage rich molecular representations including 1D SMILES strings, 2D graphs of bonds, and 3D coordinates to derive efficient machine learning kernels to address regression problems. We further expand the library of available spectral kernels for small molecules developed for classification problems to include 2.5D surface and 3D kernels using Delaunay tetrahedrization and other techniques from computational geometry, 3D pharmacophore kernels, and 3.5D or 4D kernels capable of taking into account multiple molecular configurations, such as conformers. The kernels are comprehensively tested using cross-validation and redundancy-reduction methods on regression problems using several available data sets to predict boiling points, melting points, aqueous solubility, octanol/water partition coefficients, and biological activity with state-of-the art results. When sufficient training data are available, 2D spectral kernels in general tend to yield the best and most robust results, better than state-of-the art. On data sets containing thousands of molecules, the kernels achieve a squared correlation coefficient of 0.91 for aqueous solubility prediction and 0.94 for octanol/water partition coefficient prediction. Averaging over conformations improves the performance of kernels based on the three-dimensional structure of molecules, especially on challenging data sets. Kernel predictors for aqueous solubility (kSOL), LogP (kLOGP), and melting point (kMELT) are available over the Web through: http:// cdb.ics.uci.edu.

Influence Relevance Voting: An Accurate And Interpretable Virtual High Throughput Screening Method

“…As in most chemoinformatics applications, such as the search of large databases of small molecules27-29 or the prediction of their physical, chemical, and biological properties,9,10,19,20,27 all the vHTS methods we implement depend on a quantitative notion of chemical similarity to define the local geometry of chemical space. The underlying intuition, explicitly articulated as the Similar Property Principle,30 is that the more structurally similar two molecules are, the more likely they are to have similar properties.…”

“…In both labeling schemes, the bonds are simply labeled according to their type (single, double, triple, or aromatic). In terms of graphical substructures, we consider both paths19,20 of depth d up to 2, 5, or 8 bonds, or circular substructures34 of depth d up to 2 or 3 bonds. Thus the fingerprint components index all the labeled paths, or all the labeled trees, up to a certain depth.…”

“…MAX-SIM is a particularly simple algorithm, a useful baseline for comparison. In contrast, SVMs are a highly sophisticated class of methods which have been successfully applied to other chemical classification problems19,20 and are expected to yield high performances on vHTS datasets. Comparisons against the kNN method are not included because the kNN does not rank its hits and, in prior studies, has been consistently outperformed by SVMs on chemical data.…”

Continuous Molecular Fields Approach Applied to Structure-Activity Modeling

“…A universal way of tackling the problem of molecular flexibility was suggested in paper [61] for kernel-based methods. It consists in averaging kernels over all conformations for each molecule.…”

Mean-Field Theory of Meta-learning

“…Typically ML algorithms are divided into several classes: 1) supervised learning (generated a function that maps input data into desired outputs); 2) unsupervised learning (model a set of inputs, where no prior classification is given); 3) semi-supervised learning (generate an appropriate function or classifier); 4) reinforcement learning (learn how to act given an observation of the world, where every action has some impart in the environment, with feedback of it back to the algorithm); 5) transduction (predicts new outputs based on training inputs, outputs and test inputs); and 6) learning to learn (learns its own inductive bias based on previous experience) [1]. Different algorithms of ML have been applied successfully to solve real-life problems, for example in the context of bioinformatics [2][3][4][5][6] or chemoinformatics problems [7][8][9][10][11][12].…”