Acyclic Subgraph Based Descriptor Spaces for Chemical Compound Retrieval and Classification

Abstract

In recent years the development of computational techniques that build models to correctly assign chemical compounds to various classes or to retrieve potential drug-like compounds has been an active area of research. These techniques are used extensively at various phases during the drug development process. Many of the best-performing techniques for these tasks, utilize a descriptor-based representation of the compound that captures various aspects of the underlying molecular graph's topology. In this paper we introduce and describe algorithms for efficiently generating a new set of descriptors that are derived from all connected acyclic fragments present in the molecular graphs. In addition, we introduce an extension to existing vector-based kernel functions to take into account the length of the fragments present in the descriptors. We experimentally evaluate the performance of the new descriptors in the context of SVM-based classification and ranked-retrieval on 28 classification and retrieval problems derived from 17 datasets. Our experiments show that for both the classification and retrieval tasks, these new descriptors consistently and statistically outperform previously developed schemes based on the widely used fingerprint-and Maccs keys-based descriptors, as well as recently introduced descriptors obtained by mining and analyzing the structure of the molecular graphs.

Don’t Be Afraid of Simpler Patterns

“…Also, in the chemo-informatics community, cf. [15], it is often argued that one should take into account 3D information about the compounds in addition to the 2D graph structure. Doing this within our framework would be an interesting question for further research.…”

Efficient identification of Tanimoto nearest neighbors

“…We encode each chemical compound as a sparse frequency vector of the molecular fragments it contains, represented by GF [42] descriptors extracted using the AFGen v. 2.0 [43] program. 3 AFGen represents molecules as graphs, with vertices corresponding to atoms and edges to bonds in the molecule.…”

A Probabilistic Substructure-Based Approach for Graph Classification

“…Several approaches have been developed [18,19,20,21,22] for building efficient classifiers for this task. In most of these approaches a graph is represented using various descriptors and a classification model is built using statistical or machine learning techniques.…”

“…Han et al [22] showed that frequent closed graphs based descriptor space is a better approach than the frequent subgraphs based descriptors, and it generates typically lower dimensional feature vectors. A number of methods have been proposed in recent years using cyclic patterns [20], acyclic, path and tree fragments [19] to define the descriptor space and to generate features. With this feature based representation any classification technique can be used for the classification task.…”

“…Over the years, several innovative classification algorithms for graph-based data have been developed [18,19,20,21,22,24]. Most of the algorithms are based on the underlying assumption that the intrinsic properties of a graph are rendered by its underlying substructures (nodes, edges, paths, strongly connected components, trees, subgraphs, etc).…”