Machine Learning Predicts Degree of Aromaticity from Structural Fingerprints.

Prediction of whether a compound is "aromatic" is at first glance a relatively simple task-does it obey Hückel's rule (planar cyclic π-system with 4n + 2 electrons) or not? However, aromaticity is far from a binary property, and there are distinct variations in the chemical and biological behavior of different systems which obey Hückel's rule and are thus classified as aromatic. To that end, the aromaticity of each molecule in a large public dataset was quantified by an extension of the work of Raczynska et al. Building on this data, a method is proposed for machine learning the degree of aromaticity of each aromatic ring in a molecule. Categories are derived from the numeric results, allowing the differentiation of structural patterns between them and thus a better representation of the underlying chemical and biological behavior in expert and (Q)SAR systems.

In silico prediction of pharmaceutical degradation pathways: a benchmarking study.

Zeneth is a new software application capable of predicting degradation products derived from small molecule active pharmaceutical ingredients. This study was aimed at understanding the current status of Zeneth's predictive capabilities and assessing gaps in predictivity. Using data from 27 small molecule drug substances from five pharmaceutical companies, the evolution of Zeneth predictions through knowledge base development since 2009 was evaluated. The experimentally observed degradation products from forced degradation, accelerated, and long-term stability studies were compared to Zeneth predictions. Steady progress in predictive performance was observed as the knowledge bases grew and were refined. Over the course of the development covered within this evaluation, the ability of Zeneth to predict experimentally observed degradants increased from 31% to 54%. In particular, gaps in predictivity were noted in the areas of epimerizations, N-dealkylation of N-alkylheteroaromatic compounds, photochemical decarboxylations, and electrocyclic reactions. The results of this study show that knowledge base development efforts have increased the ability of Zeneth to predict relevant degradation products and aid pharmaceutical research. This study has also provided valuable information to help guide further improvements to Zeneth and its knowledge base.

An expert system to predict the forced degradation of organic molecules.

In this paper we describe Zeneth, a new expert computational system for the prediction of forced degradation pathways of organic compounds. Intermolecular reactions such as dimerization, reactions between the query compound and its degradants, as well as interactions with excipients can be predicted. The program employs a knowledge base of patterns and reasoning rules to suggest the most likely transformations under various environmental conditions relevant to the pharmaceutical industry. Building the knowledge base is facilitated by data sharing between the users.

Correcting ligands, metabolites, and pathways.

BACKGROUND: A wide range of research areas in bioinformatics, molecular biology and medicinal chemistry require precise chemical structure information about molecules and reactions, e.g. drug design, ligand docking, metabolic network reconstruction, and systems biology. Most available databases, however, treat chemical structures more as illustrations than as a datafield in its own right. Lack of chemical accuracy impedes progress in the areas mentioned above. We present a database of metabolites called BioMeta that augments the existing pathway databases by explicitly assessing the validity, correctness, and completeness of chemical structure and reaction information. DESCRIPTION: The main bulk of the data in BioMeta were obtained from the KEGG Ligand database. We developed a tool for chemical structure validation which assesses the chemical validity and stereochemical completeness of a molecule description. The validation tool was used to examine the compounds in BioMeta, showing that a relatively small number of compounds had an incorrect constitution (connectivity only, not considering stereochemistry) and that a considerable number (about one third) had incomplete or even incorrect stereochemistry. We made a large effort to correct the errors and to complete the structural descriptions. A total of 1468 structures were corrected and/or completed. We also established the reaction balance of the reactions in BioMeta and corrected 55% of the unbalanced (stoichiometrically incorrect) reactions in an automatic procedure. The BioMeta database was implemented in PostgreSQL and provided with a web-based interface. CONCLUSION: We demonstrate that the validation of metabolite structures and reactions is a feasible and worthwhile undertaking, and that the validation results can be used to trigger corrections and improvements to BioMeta, our metabolite database. BioMeta provides some tools for rational drug design, reaction searches, and visualization. It is freely available at http://www.cmbi.ru.nl/biometa/ provided that the copyright notice of all original data is cited. The database will be useful for querying and browsing biochemical pathways, and to obtain reference information for identifying compounds. However, these applications require that the underlying data be correct, and that is the focus of BioMeta.