Application of Kohonen Neural Networks in classification of biologically active compounds

The discrimination of interaural level difference sensitivity functions: development of a taxonomic data template for modelling

Background

A major cue for the position of a high-frequency sound source in azimuth is the difference in sound pressure levels in the two ears, Interaural Level Differences (ILDs), as a sound is presented from different positions around the head. This study aims to use data classification techniques to build a descriptive model of electro-physiologically determined neuronal sensitivity functions for ILDs. The ILDs were recorded from neurons in the central nucleus of the Inferior Colliculus (ICc), an obligatory midbrain auditory relay nucleus. The majority of ICc neurons (~ 85%) show sensitivity to ILDs but with a variety of different forms that are often difficult to unambiguously separate into different information-bearing types. Thus, this division is often based on laboratory-specific and relatively subjective criteria. Given the subjectivity and non-uniformity of ILD classification methods in use, we examined if objective data classification techniques for this purpose. Our key objectives were to determine if we could find an analytical method (A) to validate the presence of four typical ILD sensitivity functions as is commonly assumed in the field, and (B) whether this method produced classifications that mapped on to the physiologically observed results.

Methods

The three-step data classification procedure forms the basic methodology of this manuscript. In this three-step procedure, several data normalization techniques were first tested to select a suitable normalization technique to our data. This was then followed by PCA to reduce data dimensionality without losing the core characteristics of the data. Finally Cluster Analysis technique was applied to determine the number of clustered data with the aid of the CCC and Inconsistency Coefficient values.

Results

The outcome of a three-step analytical data classification process was the identification of seven distinctive forms of ILD functions. These seven ILD function classes were found to map to the four “known” ideal ILD sensitivity function types, namely: Sigmoidal-EI, Sigmoidal-IE, Peaked, and Insensitive, ILD functions, and variations within these classes. This indicates that these seven templates can be utilized in future modelling studies.

Conclusions

We developed a taxonomy of ILD sensitivity functions using a methodological data classification approach. The number and types of generic ILD function patterns found with this method mapped well on to our electrophysiologically determined ILD sensitivity functions. While a larger data set of the latter functions may bring a more robust outcome, this good mapping is encouraging in providing a principled method for classifying such data sets, and could be well extended to other such neuronal sensitivity functions, such as contrast tuning in vision.

Nonlinear dimensionality reduction and mapping of compound libraries for drug discovery

Visualization of 'chemical space' and compound distributions has received much attraction by medicinal chemists as it may help to intuitively comprehend pharmaceutically relevant molecular features. It has been realized that for meaningful feature extraction from complex multivariate chemical data, such as compound libraries represented by many molecular descriptors, nonlinear projection techniques are required. Recent advances in machine-learning and artificial intelligence have resulted in a transfer of such methods to chemistry. We provide an overview of prominent visualization methods based on nonlinear dimensionality reduction, and highlight applications in drug discovery. Emphasis is on neural network techniques, kernel methods and stochastic embedding approaches, which have been successfully used for ligand-based virtual screening, SAR landscape analysis, combinatorial library design, and screening compound selection.

Assessing the Predictive Power of Unsupervised Visualization Techniques to Improve the Identification of GPCR-Focused Compound Libraries

Principal component analysis and self-organizing maps (SOMs) were compared to cluster and visualize the chemical space of a large and diverse data set. The data set comprised about 3000 G-protein-coupled receptor (GPCR) ligands for about 130 receptors and 3000 non-GPCR ligands from the World Drug Index. The molecules were described with a topological pharmacophore point histogram descriptor and a chemical fingerprint descriptor. To assess the predictive power of the clustering, a leave-multiple-out cross validation with k nearest neighbor classification was performed. The results of the classification tests and the visualization showed a clear superiority of the SOM method. SOM correctly divided the data set into two main clusters, one for the GPCR and the other for the non-GPCR ligands. Our results suggest that a continuous GPCR-ligand space exists.

Generation of QSAR sets with a self-organizing map

A Kohonen self-organizing map (SOM) is used to classify a data set consisting of dihydrofolate reductase inhibitors with the help of an external set of Dragon descriptors. The resultant classification is used to generate training, cross-validation (CV) and prediction sets for QSAR modeling using the ADAPT methodology. The results are compared to those of QSAR models generated using sets created by activity binning and a sphere exclusion method. The results indicate that the SOM is able to generate QSAR sets that are representative of the composition of the overall data set in terms of similarity. The resulting QSAR models are half the size of those published and have comparable RMS errors. Furthermore, the RMS errors of the QSAR sets are consistent, indicating good predictive capabilities as well as generalizability.

Visual and computational analysis of structure--activity relationships in high-throughput screening data

Novel analytic methods are required to assimilate the large volumes of structural and bioassay data generated by combinatorial chemistry and high-throughput screening programmes in the pharmaceutical and agrochemical industries. Recent work in visualisation and data mining has been used to develop structure--activity relationships from such chemical-biological datasets.

Database mining applied to central nervous system (CNS) activity

A data set of 389 compounds, active in the central nervous system (CNS) and divided into eight classes according to the receptor type, was extracted from the RBI database and analyzed by Self-Organizing Maps (SOM), also known as Kohonen Artificial Neural Networks. This method gives a 2D representation of the distribution of the compounds in the hyperspace derived from their molecular descriptors. As SOM belongs to the category of unsupervised techniques, it has to be combined with another method in order to generate classification models with predictive ability. The fuzzy clustering (FC) approach seems to be particularly suitable to delineate clusters in a rational way from SOM and to get an automatic objective map interpretation. Maps derived by SOM showed specific regions associated with a unique receptor type and zones in which two or more activity classes are nested. Then, the modeling ability of the proposed SOM/FC Hybrid System tools applied simultaneously to eight activity classes was validated after dividing the 389 compounds into a training set and a test set, including 259 and 130 molecules, respectively. The proper experimental activity class, among the eight possible ones, was predicted simultaneously and correctly for 81% of the test set compounds.

Mapping of protein surface cavities and prediction of enzyme class by a self-organizing neural network

An automated computer-based method for mapping of protein surface cavities was developed and applied to a set of 176 metalloproteinases containing zinc cations in their active sites. With very few exceptions, the cavity search routine detected the active site among the five largest cavities and produced reasonable active site surfaces. Cavities were described by means of solvent-accessible surface patches. For a given protein, these patches were calculated in three steps: (i) definition of cavity atoms forming surface cavities by a grid-based technique; (ii) generation of solvent accessible surfaces; (iii) assignment of an accessibility value and a generalized atom type to each surface point. Topological correlation vectors were generated from the set of surface points forming the cavities, and projected onto the plane by a self-organizing network. The resulting map of 865 enzyme cavities displays clusters of active sites that are clearly separated from the other cavities. It is demonstrated that both fully automated recognition of active sites, and prediction of enzyme class can be performed for novel protein structures at high accuracy.

Hybrid systems for virtual screening: interest of fuzzy clustering applied to olfaction

Kohonen neural networks, also known as Self Organizing Map (SOM), offer a useful 2D representation of the compound distribution inside a large chemical database. This distribution results from the compound organization in a molecular diversity hyperspace derived from a large set of molecular descriptors. Fuzzy techniques based on the "concept of partial truth" reveal to be also a valuable tool for the direct exploitation of chemical databases or SOM. In such cases a fuzzy clustering algorithm is used. In this paper, a complete hybrid system, combining SOM and fuzzy clustering, is applied. As example, a series of olfactory compounds was selected. The complexity of such information is that a same compound may exhibit different odors. It is shown how fuzzy logic helps to have a better understanding of the organization of the compounds. These hybrid systems, using simultaneously SOM and fuzzy clustering, are foreseen as powerful tools for "virtual pre-screening".