Methods

Dataset
The simplified molecular-input line-entry system (SMILES) and 3-dimensional structure data format (SDF) files of the dataset were downloaded from the PubChem Compound Database in National Center for Biotechnology Information (NCBI) and used in the further analyses. All 47 ligand molecules in the dataset, comprising 9 histamine agonists and 38 histamine antagonists, are shown in Table 1.

Structure comparison of histamine receptor ligands
The molecular similarities between histamine and other 46 chemicals were calculated from SMILES of the chemicals. The structural similarity was calculated and represented as the Tanimoto distance of each molecule from histamine. The Tanimoto coefficient for pairwise comparison of molecules is the most widely used measure of molecular structural similarity. This coefficient is defined as Tc = Nab/(Na + Nb - Nab), with Nab being the number of common bits, Na the unique bits in molecule a, and Nb the unique bits in molecule b, using a molecular fingerprint [13]. In this study, the molecular similarity was calculated as the Tanimoto coefficient using the 38-bit set.

Geometry optimization and calculation of molecular vibrational frequency
In order to calculate molecular vibrational frequency, the structure of a chemical must first be geometrically optimized. Since each provided theoretical 3-D conformer SDF is not at an energy minimum and may not represent the lowest energetic form in a vacuum, solvent, or a binding pocket, each SDF file of a ligand molecule underwent conversion to a single low-energy conformation using the general atomic and molecular electronic structure system (GAMESS) program package [14]. Restricted Hartree-Fock (RHF) calculations using Becke's exchange and Lee-Yang-Parr's correlation functionals (BLYP) density functional theory (DFT) method with 6-31G basis set were performed to optimize the geometries of the molecules. Each result was taken as the representative conformation of the molecule, although the calculation of molecular vibrational frequency has some dependence on conformation. Each geometry optimization result was subjected to the calculation step for the vibrational frequency with RUNTYP of HESSIAN in the GAMESS program.

Hierarchical clustering of the corralled intensity of molecular vibrational frequency (CIMVF)
For the simplified molecular comparison, the calculated vibrational frequencies of a molecule were then sorted in increasing order and taken into the corrals, the step size of which was 5. The intensities of each frequency in the same corral were summed up as the representative of the corral in the frequency range of 0-5,000 cm-1. As a final outcome, this potential molecular descriptor of each molecule was displayed in a 1-dimensional vector containing 1,000 elements. Finally, the similarity matrix, comprising the descriptors of 47 ligands of histamine receptor, was then subjected to hierarchical clustering in the agglomerative manner. In this study, the similarity matrix was finally clustered to make an unrooted tree of 47 vertices. The calculations of CIMVF were performed by in-house scripts, written in Python.