Aims: The aim of this letter is to explore the influence of adding hydrogen atoms to the
crystallographic structures of HIV-1 protease complexes with a series of inhibitors on the performance
of radial distribution function based descriptors recently introduced in chemoinformatic studies.
Background: Quite recently the successful application of molecular descriptors based on a radial
distribution function to correlate it with biologically interesting properties of a ligand – enzyme
complex was demonstrated. Except its predictive power, the analysis of atoms with dominant contributions
to the RDFs can be used to identify relevant atoms and interactions. Since original paper
was published on dataset consisting of the X-ray structures of complexes without hydrogen atoms,
we wonder weather addition of light atoms can provide us new piece of information.
Objective: The primarily objective is to create the model correlating the RDF based descriptors
and physicochemical properties of the HIV-1 protease complexes with inhibitors with hydrogen
atoms. Then, we will compare the performance of new model with previous one, where the hydrogen
atoms were discarded. Information about interactions between the enzyme and the inhibitors
will be extracted from the analysis of the RDF.
Methods: The radial distribution function descriptor weighted by the number of valence shell electrons
has proven to be sensitive to the changes in the structure of the enzyme and enzyme-ligand
complexes. For each structure in our data set, RDF will be calculated and using multiple linear regression
method the mathematical model will be designed correlating RDF based descriptors and
the physicochemical properties. Statistical analysis of the atom’s contribution to the total RDF will
reveal relevant interactions.
Results: The applicability of RDF based descriptor for the correlation with pKi and EC50 values is
demonstrated, while simple models containing only two or three parameters are able to explain 78
and 86 % of the variance, respectively. The models with explicitly included hydrogens are of
comparable quality with the previous models without hydrogens. The analysis of the atom’s dominant
contributions highlighted the importance of the hydroxyl groups of the inhibitor near the
Asp25 and Asp25’ residues when it is bounded to the protease.
Conclusion: Models based on the RDF weighted by the number of valence shell electrons for correlating
small number of molecular descriptors and physicocehmical properties for structures with
and without hydrogens are of comparable quality and both can be used for identification of relevant
functional groups and interactions.
Other: Our approach can be integrated to the next generation virtual screening methods, because is
fast, reliable with high predictability potential.