Construction Of An Anti-aging Compounds Database And Activity Prediction Study

Population aging is a major medical,social and economic challenge worldwide.The development of drugs that can delay aging and improve health is an important goal for the future society and a major topic in aging research.Using artificial intelligence to discover anti-aging compounds prior to the experimentation can significantly improve the efficiency and save cost.However,the application of artificial intelligence is severely hampered by the lack of high-quality data.Existing anti-aging databases suffer from slow update speed,small number of negative records and,most importantly,lack of compound structures.Therefore,we developed a highquality Anti-aging Compounds Data Base named ACDB in this study.By integrating the existing anti-aging compounds databases,including Age Fact DB,Geroprotectors,Drug Age,and Aging Atlas,as well as conducting a keyword search for " compound AND(lifespan OR life span)" in the Pub Med database,a total of 689 scientific articles were obtained.We manually extracted multidimensional information on activity data from the689 published scientific articles,including compound or crude extract names,drug doses and concentrations,animal species,strains,gender,control group mean or median survival time,treatment group mean or median survival time,changes in mean or median survival time between control and treatment groups,maximum survival time in control group,maximum survival time in treatment group,changes in maximum survival time between control and treatment groups,statistical significance(p<0.05)of changes in mean or median survival time between control and treatment groups,and literature sources.The chemical information of the compound was obtained by Pub Chem Py and Pub Chem and integrated with experimental data.A total of49 species,2585 compounds or crude extracts and 7085 experimental data were included in the current version of the ACDB.The functional design of the ACDB database mainly includes text-based data search,chemical structure-based data search,data browsing,data upload,and data download.The ACDB can provide important support for the artificial intelligence research and development of anti-aging drugs.The ACDB datasets were divided into active and inactive datasets based on drug doses and concentrations,changes in mean or median survival time between control and treatment groups,and statistical significance.We used the Synthetic Minority Oversampling Technique to augment the minority class dataset and built support vector machine and random forest prediction models named Anti-aging.We evaluated the generalization performance of the models with10 rounds of 5-fold cross-validation on the training and validation sets,using the mean AUC as the performance metric.Results from accuracy,precision,recall,F1 score,AUC,and external dataset validation consistently showed that both prediction models we constructed exhibited excellent performance.In addition,we utilized the Tanimoto coefficient to design a method for calculating the similarity between the FP2 molecular fingerprint patterns of all candidate compounds and compounds in the ACDB database,in order to evaluate the uniqueness and novelty of the candidate compounds’ structures.Using the constructed anti-aging model and the Lipinski’s rule of five,we systematically screened all approved drugs in Drug Bank and all compounds in the TCMSP.We identified 12 structurally novel and potentially anti-aging active candidate compounds with high biological utilization efficiency from Drug Bank,and 15 from TCMSP.Overall,the ACDB anti-aging compound database developed in this study provides rich and reliable data information for anti-aging compounds that have been experimentally validated.By utilizing reasonable data processing methods and artificial intelligence algorithms,the ACDB database is expected to provide reference and assistance for the discovery of anti-aging compounds.