FULLY CONNECTED CONVOLUTIONAL CORRELATION NEURAL NETWORK BASED ON THE ENERGY SPECTRUM OF MULTIPLE DOCKING: A NEW METHODOLOGY FOR SEARCHING BIOLOGICALLY ACTIVE SUBSTANCES

P.M. Vassiliev, A.V. Golubeva, M.A. Perfilev, A.N. Kochetkov

DOI: 10.22138/2500-0918-2025-22-3-279-293

Federal State Budgetary Educational Institution of Higher Education
«Volgograd State Medical University» of the Ministry of Health of the Russian Federation,
Volgograd, Russian Federation

Abstract. The purpose of the study. Development of a new architecture of a fully connected convolutional neural network based on the correlation convolution of multiple docking energy spectra for in silico search for biologically active compounds. Materials. Verified data on structure and antimicrobial activity against S. aureus of 254 known compounds structurally similar to the studied quinazolinone scaffold; validated 3D model of S. aureus peptide deformylase. Methods. A new developed algorithm for constructing a fully connected convolutional correlation neural network based on multiple docking energy spectra. The MSite program was used to construct spaces for multiple docking on a 3D model of peptide deformylase. The AutoDock Vina program was used to perform ensemble multiple docking of known S. aureus antimicrobial compounds to peptide deformylase, and their multiple docking energy spectra were formed. The FCCorNet program was used to construct a fully connected convolutional correlation neural network on the obtained data and calculate the energy values of this neural network. The AutoDock Vina program was used to perform ensemble simple docking of known S. aureus antimicrobial compounds to a specific binding site of peptide deformylase. The dependences of the expressed antimicrobial activity of chemical compounds on the energies of a fully connected neural network and on the energies of simple docking in the binding site of S. aureus peptide deformylase were studied using the methods of one-factor dispersion, discriminant, cluster and ROC analyses. Results. An algorithm for constructing a fully connected convolutional correlation neural network based on multiple docking energy spectra has been developed. 27 spaces for multiple docking have been constructed on a valid 3D model of S. aureus peptide deformylase. Ensemble multiple docking of 254 known S. aureus antimicrobial compounds into 27 spaces of peptide deformylase has been performed. A fully connected convolutional correlation neural network was constructed, reflecting the dependence of the expressed antimicrobial activity of chemical compounds on the energy of this network. Using four methods of multivariable statistical analysis, a study was conducted of the neural network constructed for peptide deformylase and its accuracy was tested in comparison with the results of using simple docking. Very high statistical reliability p=5.37·10-14 of using the developed methodology for assessing the affinity of ligands to a biotarget was shown. In all respects, the accuracy of searching for active compounds using the energy of a fully connected convolutional correlation neural network based on multiple docking exceeds the accuracy of the model built on the basis of the energy of simple docking into a specific binding site. Conclusions. A new artificial intelligence method was developed for in silico searching biologically active substances based on a fully connected convolutional correlation neural network and multiple docking energy spectra.

Keywords: artificial intelligence, biological active compounds, fully connected convolutional neural network, multiple docking energy spectrum, correlation convolution

There is no conflict of interest.
Contact information of the author responsible for correspondence:
Pavel M. Vassiliev
pvassilev@mail.ru
Recieved 28.04.2025
For citation: Vassiliev P.M., Golubeva A.V., Perfilev M.A., Kochetkov A.N. Fully connected convolutional correlation neural network based on the energy spectrum of multiple docking: a new methodology for searching biologically active substances. [Online] Vestn. Ural. Med. Akad. Nauki. = Journal of Ural Medical Academic Science. 2025, Vol. 22, no. 3, pp. 279–293. DOI: 10.22138/2500-0918-2025-22-3-279-293 (In Russ)

REFERENCES
1. Sarkar C., Das B., Rawat V.S., Wahlang J.B., Nongpiur A., Tiewsoh I., Lyngdoh N.M., Das D., Bidarolli M., Sony H.T. Artificial Intelligence and Machine Learning Technology Driven Modern Drug Discovery and Development. International Journal Molecular Sciences. 2023; 24(3): 2026. DOI: 10.3390/ijms24032026.
2. Leijnen S., Van Veen F. The Neural Network Zoo. Proceedings. 2020; 47(4): 9. DOI: 10.3390/proceedings2020047009,
3. Vassiliev P.M., Kochetkov A.N., Spasov A.A., Perfilev M.A. Multiple docking energy spectrum as a multidimensional metric of chemical compounds affinity to pharmacologically relevant biotargets. Volgograd Scientific Medical Journal. 2021; 3: 57-61. https://www.volgmed.ru/uploads/journals/articles/1636976502-bulletin-2021-3-4019.pdf. [Vasil'yev P.M., Kochetkov A.N., Spasov A.A., Perfil'yev M.A. Spektr energiy mnozhestvennogo dokinga kak mnogomernaya metrika affinnosti khimicheskikh soyedineniy k farmakologicheski relevantnym biomishenyam. Volgogradskiy nauchno-meditsinskiy zhurnal. 2021; 3: 57-61. https://www.volgmed.ru/uploads/journals/articles/1636976502-bulletin-2021-3-4019.pdf.]. (In Rus.).
4. Sangshetti J.N., Kalam Khan F.A., Shinde D.B. Peptide deformylase: a new target in antibacterial, antimalarial and anticancer drug discovery. Current Medicinal Chemistry. 2015; 22(2): 214-236. DOI: 10.2174/0929867321666140826115734.
5. Fieulaine S., de Sousa R.A., Maigre L., Hamiche K., Alimi M., Bolla J.-M., Taleb A., Denis A., Pages J.-M., Artaud I., Meinnel T., Giglione C. A unique peptide deformylase platform to rationally design and challenge novel active compounds. Scientific Reports. 2016; 6(1): 35429. DOI: 10.1038/srep35429.
6. Lee S.J., Lee S.-J., Lee S.K., Yoon H.-J., Lee H.H., Kim K.K., Lee B.J., Lee B.I., Suh S.W. Structures of Staphylococcus aureus peptide deformylase in complex with two classes of new inhibitors. Acta Crystallographica Section D — Structural Biology. 2012; 68(7): 784-793. DOI: 10.1107/S0907444912011912.
7. Che X., Hu J., Wang L., Zhu Z., Xu Q., Lv J., Fu Z., Sun Y., Sun J., Lin G., Lu R., Yao Z. Expression, purification, and activity assay of peptide deformylase from Escherichia coli and Staphylococcus aureus. Molecular and Cellular Biochemistry. 2011; 357(1-2): 47-54. DOI: 10.1007/s11010-011-0874-6.
8. Peyrusson F., Butler D., Tulkens P.M., Van Bambeke F. Cellular pharmacokinetics and intracellular activity of the novel peptide deformylase inhibitor GSK1322322 against Staphylococcus aureus laboratory and clinical strains with various resistance phenotypes: studies with human THP-1 monocytes and J774 murine macrophages. Antimicrobial Agents and Chemotherapy. 2015; 59(9): 5747-5760. DOI: 10.1128/AAC.00827-15.
9. Csizmadia F. JChem: Java applets and modules supporting chemical database handling from web browsers. Journal of Chemical Information and Modeling. 2000; 40(2):323-324. DOI: 10.1021/ci9902696.
10. Stewart J.J.P. Optimization of parameters for semiempirical methods VI: more modifications to the NDDO approximations and re-optimization of parameters. Journal of Molecular Modeling. 2013; 19(1): 1-32. DOI: 10.1007/s00894-012-1667-x.
11. Laskowski R.A., Swindells M.B. LigPlot+: multiple ligand-protein interaction diagrams for drug discovery. Journal of Chemical Information and Modeling. 2011; 51(10): 2778-2786. DOI: 10.1021/ci200227u.
12. Dallakyan S., Olson A.J. Small-molecule library screening by docking with PyRx. Methods in Molecular Biology. 2015; 1263: 243-250. DOI: 10.1007/978-1-4939-2269-7_19.
13. Trott O., Olson A.J. AutoDock Vina: improving the speed and accuracy of docking with a new scoring function, efficient optimization and multithreading. Journal of Computational Chemistry. 2010; 31(2): 455-461. DOI: 10.1002/jcc.21334.
14. Hilbe J.M. Statistica 7: an overview. The American Statistician. 2007; 61(1): 91-94. DOI: 10.1198/000313007X172998.
15. Hanley J.A., McNeil B.J. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology. 1982; 143(1): 29-36. DOI: 10.1148/radiology.143.1.7063747.
16. Hebb D.O. Organisation of Behavior. New York: John Wiley & Sons; 1949.
17. Hopfield J.J. Neural networks and physical systems with emergent collective computational abilities. Proceedings of the National Academy of Sciences USA. 1982; 79(8): 2554-2558. DOI: 10.1073%2Fpnas.79.8.2554.
18. Glotov N.V., Zhivotovsky L.A., Khovanov N.V., Khromov-Borisov N.N. Biometrics. Leningrad: Publishing House of Leningrad University; 1982. [Glotov N.V., Zhivotovskiy L.A., Khovanov N.V., Khromov-Borisov N.N. Biometriya. L.: Izd-vo Leningr. un-ta; 1982.]. (In Rus.).
19. Vasil’ev P.M., Kochetkov A.N., Yanalieva L.R., Vorfolomeeva V.V., Babkova V.A., Butov G.M., Burmistrov V.V., Popov O.A. Hierarchical Assessment of the Structural Similarity of Pharmacologically Active Compounds. Pharmaceutical Chemistry Journal. 2021; 54(10): 1019-1023. DOI: 10.1007/s11094-021-02314-2.
20. Kim J.-O., Mueller C.W., Klekka W.R., et al. Factor, discriminant and cluster analysis. Moscow: Finance and Statistics; 1989. [Kim Dzh.-O., M'yuller CH.U., Klekka U.R. i dr. Faktornyy, diskriminantnyy i klasternyy analiz. M.: Finansy i statistika; 1989.]. (In Russ.).
21. Ahrens H., Leuter J. Multivariate Analysis of Variance. Moscow: Finance and Statistics; 1985. [Arens KH., Loyter YU. Mnogomernyy dispersionnyy analiz. M.: Finansy i statistikal; 1985.]. (In Russ.).
22. Gupta M., Sharma R., Kumar A. Docking techniques in pharmacology: How much promising? Computational Biology and Chemistry. 2018; 76: 210-217. DOI: 10.1016/j.compbiolchem.2018.06.005.

Authors
Pavel M. Vassiliev
Doctor of Biological Sciences, senior researcher at the Higher Attestation Commission (Associate Professor), Head of Laboratory for Information Technology in Pharmacology and Computer Modeling of Drugs for Research Center of Innovative Medicines with Pilot Industrial Production
pvassilev@mail.ru

Arina V. Golubeva
Junior researcher of Laboratory for Information Technology in Pharmacology and Computer Modeling of Drugs for Research Center of Innovative Medicines with Pilot Industrial Production
arina_arina_golubeva@mail.ru

Maksim A. Perfilev
Junior researcher of Laboratory for Information Technology in Pharmacology and Computer Modeling of Drugs for Research Center of Innovative Medicines with Pilot Industrial Production
maxim.firu@yandex.com

Andrey N. Kochetkov
System administrator of Laboratory for Information Technology in Pharmacology and Computer Modeling of Drugs for Research Center of Innovative Medicines with Pilot Industrial Production
akocha@mail.ru

Federal State Budgetary Educational Institution of Higher Education «Volgograd State Medical University» of the Ministry of Health of the Russian Federation
Volgograd, Russian Federation