The Pharmacophore Kernel for Virtual Screening with Support Vector Machines

Mahé, Pierre; Ralaivola, Liva; Stoven, Véronique; Vert, Jean-Philippe

doi:10.1021/ci060138m

“…In our experience, these more complex kernels applied to predicted structures are still outperformed by the best 2D kernels, as shown in several additional experiments that were carried on the four data sets used by Mahé et al 23 for classification problems (Table 7). One exception is the alkane boiling point data set where the 3D contact histogram kernels yield the best predictions.…”

Section: Discussionmentioning

confidence: 74%

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

Azencott

¹

,

Ksikes

²

,

Swamidass

³

et al. 2007

J. Chem. Inf. Model.

Self Cite

View full text Add to dashboard Cite

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.

show abstract

“…For instance, in the case of k ) 3, we can use a pharmacophore representation, whereby a molecule is represented by the list of all of its triplets of atoms (or even groups of atoms), with the pairwise distances between the pairs of atoms in each triplet, and the corresponding labels, which, beyond atom type, can include information about size, polarity, electronegativity, and so forth. This approach is the same as the one recently described in Mahé et al, 23,24 where the authors use a labeling scheme based on the Morgan indices 25,26 that increase the specificity of the labels by including topological information about adjacent atoms. If desirable, a more compact representation is derived by building histograms for each class of triplets (e.g., C-C-C, C-C-O) on the basis of the size of the smallest sphere that contains all three points (or the largest pairwise distance in the triplet).…”

Section: D Kernels Based On Atomic Coordinates and Pharmacophoresmentioning

confidence: 99%

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

Azencott¹,

Ksikes²,

Swamidass³

et al. 2007

Self Cite

View full text Add to dashboard Cite

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.

show abstract

“…Actually, in some domains where the number and/or the diversity of the available examples are limited, as in the domain of chemometry [12], one might learn average properties, these might do well on the test set, and still be poorly related to the target concept; some evidence for the possibility of such a phenomenon was presented in [1], where the test error could be 2% or lower although the concept learned was a gross overgeneralization of the true target concept.…”

Section: Discussion and Perspectivesmentioning

confidence: 99%

“…Besides early approaches [4], specific kernels were designed for MIP problems [5,3,12,11]. The basic idea is to define the kernel K of two bags of instances as the average of the kernels k between their instances:…”

Section: State Of the Artmentioning

confidence: 99%

See 1 more Smart Citation

A Phase Transition-Based Perspective on Multiple Instance Kernels

Gaudel

¹

,

Sebag

²

,

Cornuéjols

³

Inductive Logic Programming

View full text Add to dashboard Cite

Abstract. This paper is concerned with Relational Support Vector Machines, at the intersection of Support Vector Machines (SVM) and Inductive Logic Programming or Relational Learning. The so-called phase transition framework, primarily developed for constraint satisfaction problems (CSP), has been extended to relational learning, providing relevant insights into the limitations and difficulties thereof. The goal of this paper is to examine relational SVMs and specifically Multiple Instance Kernels along the phase transition framework; a specific CSP formalization for multiple instance problems, inspired by chemometry applications, is proposed. Ample empirical evidence based on a set of order parameters shows the existence of an unsatisfiability region for standard MIP-SVM approaches. A statistical analysis for these findings is proposed, establishing a lower bound of the generalization error depending on the satisfiability probability.

show abstract

The Pharmacophore Kernel for Virtual Screening with Support Vector Machines

Cited by 76 publications

References 36 publications

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

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

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

A Phase Transition-Based Perspective on Multiple Instance Kernels

Contact Info

Product

Resources

About