Current Computer - Aided Drug Design - Volume 1, Issue 2, 2005

QSAR and subsequent molecular design are very important steps in drug discovery. Through QSAR, one derives a model that relates a set of molecular descriptors to a biological activity. The resulting model can be used to predict the activity values of new compounds in molecular design. QSAR models range from simple, parametric equations to complex, non-linear models. These models have each specific advantage and shortcoming derived from their own algorithms. We have developed hybrid approaches combining GA, multiway PLS and NN to utilize specific advantage and to cover specific shortcoming of each method. We have picked up five topics and outlined with the representative examples in this review article.

This article is presenting the most recent trends in the ways that the bioactive molecules of the future will be designed. They rely on important recent discoveries for our understanding of the global organization of the cell, giving evidence that biological networks form small-world and scale-free structures. These networks are composed of well-defined modules, with nodes connected by relatively short paths that allow for fast signaling. The few very connected nodes, that are unlikely to be affected by random alterations, support the proper functioning of the whole system. The fact that in cells everything is connected to everything explains why monogenic diseases, associated to the alteration of individual genes, were found to be an exception rather than a rule. The newly developed chemogenomic technology is offering an alternative to the traditional animal-centric pharmacological approach in the need to evaluate bioactive molecules efficacy on intact biological systems, where the multiple targets and pathways reside in their natural environment. The existence of regulatory and interaction 'neural centres' or hubs in these networks is setting new perspectives to target identification and validation. With these new technologies at hand, we are entering an exciting new era where the pharmacological targets will shift from single proteins, to functional protein complexes, to whole networks determining precise cellular states, and where the new cures and foods will be no more based on single active ingredients but will represent molecular cocktails or multiple ligands with components targeting the neural centers of whole disease-associated molecular networks.

A major problem in traditional quantitative structure-activity relationships (QSARs) analysis is to select suitable chemical descriptors from a large pool of variables. Decisions against or in favor of a particular descriptor depends entirely on the result of statistically based hypothesis testing. Uncertain results may be produced in presence of multicollinear descriptors and flagged observations (high-leverage points, outliers, influential data). To satisfy the assumptions for hypothesis testing, diagnostic statistics and subsequent design repair are employed. Here we show an example with nonnucleoside HIV-1 reverse transcriptase inhibitors.

The mechanisms behind ADME (absorption, distribution, metabolism, and excretion) related properties and toxicity endpoints are usually complex, and many are not fully understood. As a result, most ADMET predictive models are not based on theoretical principles, but are derived from experimental data. ADMET properties are best analyzed by projecting them onto the compounds of the training set. There are multiple advantages to projecting the ADMET properties from the problem domain to the chemical domain. Projection simplifies the problem, and avoids the entanglement of needing to invoke specific mechanisms. Projection focuses on the most important, and most tractable, aspect of the problem -- the related properties of the compounds themselves. In this review article, the general requirements of the chemical space to be projected are discussed, including the size and diversity of the training set and the accuracy of the biological measurements, and the process is illustrated using an analogue of a real projection. Also, the successes and pitfalls of the projection method in recent ADMET predictions are reviewed.

Wider acceptance of QSARs would result in a constellation of benefits and savings to both private and public sectors. For this to occur, particularly in regulatory applications, a model's limitations need to be identified. We define a model's limitations as encompassing assessment of overall prediction accuracy, applicability domain and chance correlation. A general guideline is presented in this review for assessing a model's limitations with emphasis on and examples of application with consensus modeling methods. More specifically, we discuss the commonalities and differences between external validation and cross-validation for assessing a model's limitations. We illustrate two common ways of assessing overall prediction accuracy, depending on whether or not the intended application domain is predefined. Since even a high quality model will have different confidence in accuracy for predicting different chemicals, we further demonstrate using the novel Decision Forest consensus modeling method a means to determine prediction confidence (i.e., certainty for an individual chemical's prediction) and domain extrapolation (i.e., the prediction accuracy for a chemical that is outside the chemistry space defined by the training chemicals). We show that prediction confidence and domain extrapolation are related measures that together determine the applicability domain of a model, and that prediction confidence is the more important measure. Lastly, the importance of assessing chance correlation is emphasized, and illustrated with several examples of models having a high degree of chance correlations despite cross-validation indicating high prediction accuracy. Generally, a dataset with a skewed distribution, small data size and/or low signal/noise ratio tends to produce a model with high chance correlation. We conclude that it is imperative to assess all three aspects (i.e., overall accuracy, applicability domain and chance correlation) of a model for the regulatory acceptance of QSARs.

Development of computer methods in molecular biology and fast growth of microbial genomics data enabled new approach based on selecting in silico antigenic components to design vaccine constructs. It is expected that application of this technology will eliminate side effects of new vaccines and reduce the time consumption and financial expenses. The bioinformatics methods of sequence analysis are used to reveal the most prospective proteins or protein fragments of infectious agents as candidates for vaccine design. In these studies the specialized molecular immunology databases are widely used. The new approach (“Reverse vaccinology”) could help in designing vaccines against diseases where traditional methods are not successful, e.g. when the viral genome reveals the extreme variability and permanent changes of antigenic properties that make difficulties for selection of molecular targets for medicines and candidate vaccines. A number of informational resources are already designed to collect and provide genomic data on certain microbes or viruses. The peculiarity of such resources is presentation of data, characterizing the different genomic variants of the same infectious agents. These structural data coupled with information on functional / immune features and software tools have to compose basis for constructing a new generation of vaccines against “common” and new infections such as AIDS, Hepatitis C, and SARS. The approaches published in literature, as well as the authors' original results are discussed.