Background: Molecular structure is the basis for establishing of quantitative
structure –activity relationships (QSARs). The molecular graph as well as the simplified
molecular input line entry systems (SMILES) are possible ways to represent the
molecular structure for QSAR analysis by means of the CORAL software.
Methods: In spite of apparent influence of distribution of available data on the training set and validation set, the majority
of works dedicated to quantitative structure - property/activity relationships (QSPRs/QSARs) are based solely on one split
into visible set (i.e. the training set) and invisible set (i.e. the validation set). We deem that each QSAR approach should
be estimated for a group of splits into the training set and validation set. The use of this principle for the case of antituberculosis
agents is the essence of this work.
Results: Ten splits of the data on anti-tuberculosis agents into the training and test sets have been examined. The statistical
approach to define the domain of applicability has been suggested and estimated. A collection of molecular structures
which should be very effective anti-tuberculosis agents according to the established model is suggested.
Conclusion: The CORAL software available on the Internet can be used for the QSAR analysis of other molecular structures
which are capable anti-infective agents.