Implementation of orbital graph model for partial atom charges definition in virtual screening technique
ABSTRACT
Partial atom charges are one of the key features of ligand description in molecular docking approach of virtual screening technique. We implemented partial atomic charge definition model originally proposed by Oliferenko A. A. [1] into virtual screening engine. This model represents atom orbital network via electric topology graph and calculates atom partial charges by equalization of atom’s orbital electronegativity. The accuracy of molecular docking method was compared for charges derived from orbital graph model, molecular graph model and AM1 semi empirical approach.
THEORY
Ohm's low:
where Y is the matrix of nodal conductivity, Ô is the column matrix of the potentials and J is the column matrix of the nodal currents.
The solution in matrix form is:
where D is the diagonal matrix of graph vertex, A is the adjacency matrix of the network graph and I is the identity matrix of the same order as A and D.
where X is the equalized electronegativity and X0 is the table electronegativity.
Orbital graph model:
where D is the normalization factor, q is the orbital charge, χ is the orbital electronegativity, η is the orbital hardness, E is the atomic energy, IP is the ionization potential and EA is the electron affinity.
Molecular graph model:
where m is the normalization factor, qi is the atom charge and X0 is the atom table electronegativityRESULTS
The highest accuracy was obtained for molecular graph charge definition model. The accuracy values were calculated as correlation index of docking score with natural logarithm of half of the rest ferment activity in presence of 30 mM inhibitor.
Accuracy dependence of molecular docking is a function of electrostatic coefficient (for molecular graph model).
The best correlation index for molecular graph model is 3.643e-2 (point 2.4), for orbital graph model is 3.169e-2 (point 1.0) and for AM1 semi empirical approach is 3.624e-2 (point 0.6)
SOFTWARE BACKGROUND
Implementation of orbital graph model for partial atom charges definition was developed as a part of y_system. Y_system is a developing project of full flexible receptor based virtual screening system which is directed to distributed computing. We had started this project in autumn of 2004. The core of y_system is an interaction between built-in database engine with several different calculation plug-ins. The picture below describes y_system structure:
Y_system is written in C for Linux OS. It’s main features are:
1. Integrated databases for the access to molecular complexes coordinates,
compound structures, QSAR tables etc
2. Distributed calculations support
3. Compounds preprocessing
4. Entropy-based scoring function for receptor-based virtual screening
5. 3D/4D QSAR for virtual complexes sets
6. Full-flexible molecular docking
7. Protein-protein screening and design supportTECHNICAL FACILITIES
Our laboratory has a cluster of 8 computers (P4 3.2GHz, 1Gb of memory, Linux OS) for the computations. This cluster mainly works under control of y_system program suite. Server node keeps database server y_dbase and controls all other nodes in cluster via y_kernel and y_console. In addition, every node (including server) has a calculation plug-ins connected via sockets to y_kernel (task manager of y_system).
These hardware are enough for processing about 1 000 000 compounds per month via receptor based virtual screening.
We are interesting in collaboration in the fields of virtual screening of proteins, small compounds as well as modeling algorithms development.
REFERENCES1. Oliferenko A.A., Krylenko P.V., Palyulin V.A. and Zefirov N.S. A new scheme for electronegativity equalization as a source of electronic descriptors: application to chemical reactivity. SAR and QSAR in Environmental Research, Vol. 13 (2), (2002) 297305
© YARMOLUK RESEARCH LAB