Predicting the Probable Receptor Targets for a Potential Drugs Based on the Assessment of Their Similarity With Endogenous Ligands

Todays pharmacology supposes that pharmaceutical agents interact with various physiological receptors which are the targets for neurotransmitters, hormones and other endogenous bioregulators [1]. The same physiological effects can be induced by activation/inhibition of different receptors. Contrary, interaction with several receptors can decrease the effects, induced by each receptor separately. Therefore, the prediction of probable receptor targets for substance is important for design of agent with desirable pharmacological effect.


Introduction
Todays pharmacology supposes that pharmaceutical agents interact with various physiological receptors which are the targets for neurotransmitters, hormones and other endogenous bioregulators [1].The same physiological effects can be induced by activation/inhibition of different receptors.Contrary, interaction with several receptors can decrease the effects, induced by each receptor separately.Therefore, the prediction of probable receptor targets for substance is important for design of agent with desirable pharmacological effect.We develop computer system SIMEST for multiple similarity assessment of a new compound with high selective small ligands of known receptors.The principal idea is that the similar compounds will interact with the same receptors.
The SIMEST includes: the software for similarity estimation between a pattern molecule and each of the ligands; the database of highly selective small ligands (endogenic bioregulators and their analogs).

Methods
The Similarity Estimation module is arranged as the ISIS/Base application (MDL�Information Systems, Inc.).Similarity assessment is based on in-house developed topological descriptors and widely used Tanimoto coefficient.
The Database works under the ISIS/Base and integrates the structure and activity data for both endogenous ligands and highly selective agonists and antagonists for 100 receptor subtypes.

Topological Descriptors
The structure description is based on connection The descriptor of higher level for an atom is generated iteratively and includes the descriptors of the previous level for an atom and its neighbours.This process can be continued up to any level substituting the list of nearest neighbor atoms by appropriate descriptors.
It is shown that inclusion of descriptors up to the 3rd level provides the satisfactory accuracy of recognition.These descriptors are reffered further as the Sub-Structure Descriptors (SSD).
Example of coding by SSD of 1 st and 2 nd levels for phenol are shown in Figure below.The descriptors of 3 rd level are not presented here because of their large size.
Tanimoto coefficient [2] is used to measure the similarity between two molecules A and B: where A(i) and B(i) are equal to 1 when i-th descriptor is found in molecule A and B respectively and 0 when the i-th descriptor is absent; M is the total number of descriptors in the dataset.

How SIMEST Works
With SIMEST one can input the pattern structure and find the most similar ligands from the Database.The result is presented as the list of receptor subtypes arranged in descending order of corresponding similarity coefficients.
Therefore, high similarity of the compound with particular endogenous-like ligand, that is Agonist and/or Antagonist of the Receptor, lead one to the conclusion that the compound probably has the same activity.
The figure below demonstrates the result of similarity searching for compound 206830 from MDDR 96.2 database, which have 5-HT4 Agonist, 5-HT3 Antagonist activities.

Results and Discussion
The possibilities of SIMEST to distinguish the active compounds from inactive are evaluated on the 17124 compounds from MDDR 96.2 database (MDL Information Systems, Inc.).The accuracy of active compounds recognition (for each kind of activity) is calculated as [3]: where: N{s(i, 1)>s(i, 0)} is the number of cases when the active compound is more similar to i-th active ligand than inactive one, when all pairs of active and inactive compounds are compared; N b is the numbers of active ligands in the SIMEST/Database; N 1 and N 0 are the numbers of active and inactive compounds in the evaluation set.
Using this criterion, the accuracy of active compounds recognition has been calculated for 49 different mechanisms of action.These results are given below.The average accuracy of recognition is significantly higher for agonists than for antagonists.The result is probably explained by the fact that antagonists have the less specific structures than the agonists.

Conclusions
New computer system SIMEST for assessing similarity of chemical substances with endogenous-like ligands is developed.
It is shown in experiments with MDDR database that the recognition of specific activity averages about 87% for agonists and 73% for antagonists.
table (C) and table of atoms types (AT).Connection table contains the information about bonds in the molecule.We do not specify various bond types, but take into account all hydrogens congruous to the valencies and charges of atoms.AT table includes the element types for each atom in a molecule.All chemical elements are classified according to the rules given below.