In Silico Prediction Of Elimination Half-life,Metabolic Reactions,and Analysis Of Associated Targets Of Acute Kidney Injury

Absorption,distribution,metabolism,excretion and toxicity(ADME/T)make significant effect on the drug's efficacy and safety.Predictive models for accurately assessing the ADME/T properties in early development stages are highly helpful to increase the efficiency of drug research and reduce the cost of drug development.With the accumulation of the experimental data,the promotion of the super computer,and the development of in silico methods,computer aided prediction of ADME/T properties has attracted more and more attention from the drug researchers.The great advantages of in silico predictive models are their characteristics of low cost and high throughput.In silico prediction of ADME/T also plays an important role in the guidance for rational drug use.Machine learning methods have been often applied in ADME/T studies for their low-cost and high-efficiency.In the first section of my dissertation,I introduced many applications of machine learning methods to predict ADME/T properties and discussed the challenges of the in silico prediction of ADME/T studies.Elimination half-life of drugs is the time required for the concentration of a compound to fall to half of its initial concentration,which is an important pharmacokinetic parameter.Elimination half-life determines the rate of elimination and duration of action.It is often time-consuming and costly to make experimental evaluation of half-life.Thus,it is necessary to develop in silico predictive models for elimination half-life.In the second section of my dissertation,the high-quality predictive models were built by several machine learning methods,such as gradient boosting machine(GBM),support vector regressions(RBF-SVR and Linear-SVR),local lazy regression(LLR),SA,SR,and GP.The consensus models were also explored to improve the accuracy of the results.Among these prediction models,GBM showed the best performance(R~2=0.820 and RMSE=0.555 for the test set).And then some important descriptors related to eliminaton half-life were identified and analyzed.Many drugs,having got into the human body,will undergo chemical transformations induced by several metabolic enzymes.The study of metabolism prediction is very important in many areas,including food safety,environmental studies,pharmaceutical research and so on.The study of drug metabolism is one of the most important concern in drug development and research.The pharmacokinetic and pharmacodynamic properties of drugs are highly dependent on the process of drug metabolism.It is time-consuming and costly to make prediction of drug metabolism by experimental studies.However,in silico prediction of metabolic reactions is high-speed and effective,which attracts more and more attentions from researchers.In the third section of my dissertation,many metabolic reactions were retrieved from Accelrys Metabolite Database.The sites and types of metabolic reactions were then identified by MCS algorithm and reaction rules.The fingerprint descriptors were used to characterize the molecular environment of the sites of metabolic reactions.Finally,in silico prediction models for 23 types of metabolic reactions were developed by deep neural nets.The AUC of the models are more than 0.7,showing certain predictive ability.The kidney is the most important excretion organ in human body and is vulnerable to various potential nephrotoxins,such as environmental chemicals,diagnostic agents and medications.Acute kidney injury is a clinical critical disease,and has high mortality.The molecular mechanisms of acute kidney injury are highly complex,and differ between various drugs.It is helpful to elucidate the molecular mechanisms and predict nephrotoxicity by making analysis of potential associated targets of drug-induced acute kidney injury.In the fourth chapters,we collected the positive and negative compounds of acute kidney injury from CTD and DrugBank database.And the interacting target genes/proteins of these compounds were also extracted for the next analysis.In order to examine the significance of the association between each target gene and acute kidney injury,fisher's exact test was adopted to measure the significance.Finally,these target genes were ranked by p-value.Among the top 20 target genes,14 genes were associated with kidney damages and 5 of these genes were explicitly associated with acute kidney injury according to the literature reports.Other top genes could be potential targets of acute kidney injury.