Genetic Algorithm And Deep Learning Approach Based Automatic Design Of ATR Inhibitors

Ataxia telangiectasia and Rad-3 related protein kinases（ATR）as important kinases in the PIKK family,which is not only related to gene recombination,but also play a key regulatory role in DNA damage repair pathways.Due to defects in the DNA damage response mechanism in tumor cells,which is more dependent on ATR,so that inhibition of ATR can effectively inhibit tumor cells and has less impact on healthy cells.Therefore,ATR is regarded as a valuable pharmacological target for anticancer therapy due to its indispensable bioactivities in multiple cancer cells.However,the number of known ATR inhibitors is rare,and there are no clinical drugs on the market.It is still a huge challenge to develop novel and specific ATR inhibitors as clinical candidates.This work attempts to combine genetic algorithm with deep learning model to construct an automatic design system for ATR inhibitors,efficiently design and obtain ATR inhibitors with diverse structures,and verify their synthesis and anti-proliferative activity of tumor cells in a dose-dependent manner.This study is mainly divided into the following five parts:PartⅠ.IntroductionThis chapter reviews the structure and physiological functions of ATR,the development of known ATR inhibitors,computer-aided drug design strategies,and the research progress of deep learning and genetic algorithms in the field of biomedicine,so as to provide theoretical foundation and technical support for the research of this subject.PartⅡ.Construction of ATR inhibitor design system based on genetic algorithm and deep learningIn this chapter,889 effective ATR inhibitors were collected from the Binding DB database.115 kinds of molecular descriptors and ECFP molecular fingerprints were calculated using the RDKit toolkit.Based on the recursive feature elimination method,84 kinds of molecular descriptor with high correlation of IC₅₀ were obtained.Furthermore,three kinds of deep learning regression prediction models with molecular descriptors（model 1）,molecular fingerprints（model 2）,combined molecular descriptors and molecular fingerprints（model 3）as input data were constructed respectively,used grid search algorithm to optimize the performance of the regression model,and through its performance comparison,the model 3 with best performance（R²=0.755,RMSE=0.484）is finally used as the fitness evaluation approach of the genetic algorithm.Secondly,based on the Python scripting language,an effective combination of genetic algorithms and deep learning regression prediction model（model 3）has been developed to construct an ATR inhibitor automatic design system,which is characterized by the ability to efficiently design a series of ATR inhibitors with diverse structures.The initial population of the system was set to 100 SMILES of the best active ATR inhibitors,the gene mutation probability was 0.01,and the evolutionary generation number was 1000,and finally 195 novel ATR inhibitors were automatically generated.PartⅢ.Evaluation of ATR inhibitorsIn this chapter,molecular docking technology is used to evaluate the binding ability of the 195 novel ATR inhibitors to the target.The results showed that all 195 ATR inhibitors can bind to the active site of the target.Secondly,based on an algorithm to evaluate the difficulty of compound synthesis,the feasibility of 195 novel ATR inhibitors was investigated.The calculation results showed that the synthetic feasibility scores of 195 ATR inhibitors are all less than 5,indicating that the designed compound meets the laboratory synthesis conditions.In the end,with the number of hydrogen bond interactions andπ-stacking interactions more than 3 as the selection criteria,four hits（compounds 2,9,30 and 176）were selected from 195 novel ATR inhibitors for further analysis.PartⅣ.Chemical synthesis of ATR inhibitorsThis chapter aims to chemically synthesize the designed novel ATR inhibitors in order to obtain molecular entities and provide a basis for further experimental evaluation of anti-proliferation activity of tumor cells.First,we designed and optimized the synthetic route of four hits according to the retro-synthetic analysis approach.Second,four hits were synthesized according to the aboratory conditions.Finally,the structure of the ATR inhibitors was characterized by means of ¹H NMR,¹³C NMR and MS analysis methods.PartⅤ.Study on the anti-tumor proliferation activity of ATR inhibitorsThis chapter used human breast cancer cell line（MCF-7）,doxorubicin hydrochloride（DOX） as a control drug,using 4 novel hits alone and using two in combination with doxorubicin hydrochloride at a molar ratio of 1:1.The cellular dosing regimen evaluates the anti-proliferation activity of four hits.The results showed that the four hits all have anti-proliferation activity of MCF-7 cell line,but the effect of the single drug is not ideal,with IC₅₀ of 332.211μM,179.750μM,304.426μM,and 318.141μM,respectively.In addition,the combined administration of a new type of ATR inhibitor and DOX has a synergistic anti-proliferation effect.The synergistic effect of compound 2 combined with DOX is CI=0.515,exerting a highly synergistic effect.The synergistic effect of compound 9 combined with DOX is CI=0.334,exerting a strong synergistic effect.The synergistic effect of compound 30 and DOX is CI=0.777,which exerts a moderate synergistic effect.The synergistic effect of compound 176 combined with DOX is CI=0.505,which exerts a highly synergistic effect and can significantly enhance the anti-proliferation activity.In summary,the automatic design system for ATR inhibitors constructed in this study can efficiently design a series of potential compounds with diverse structures,which is expected to provide data support for clinical studies of ATR inhibitors.In addition,the computer-aided drug design strategy adopted in this work can provide technical support and theoretical foundation for the development of other drugs.