Drug Organ Toxicity Evaluation In Basis Of Real-World Big Data

Drug toxicity is estimated to be responsible for about 1/3 of new drug failure;it is also a key component in nowadays precision medicine.Adverse drug reactions(ADRs)in different organs are the direct clinical manifestations of drug organ toxicity.However,quantitative evaluation of clinical organ toxicity in a comprehensive,unbiased,and high-throughput manner is an unresolved problem yet.The severity of ADR contains fruitful information of drug toxicity degree and its organ targeting preference.Therefore,ADR severity will serve as one of the key factors for the comprehensive evaluation of drug organ toxicity.In Chapter Two of this thesis,we designed an improved Discounted Cumulative Gain(IDCG)model to quantitatively evaluate the severity of 6,277 ADRs induced by 774 drugs via mining 11,853,341 realworld adverse drug event(ADE)reports derived from the US FDA’s Adverse Event Reporting System(FAERS).The ADRs were further graded according to their severity scores,and the grades were compared with Common Terminology Criteria for Adverse Events(CTCAE,v5.0).The results demonstrated that 72.16%of ADR severity assigned by the model were consistent with that determined by the experts.In Chapter Three of the thesis,we constructed a Linear Weighted Summation(LWS)model to comprehensively evaluate the broad-spectrum ADR profile of 774 drugs and their organ toxicity by incorporating the quantified ADR severity and frequency with 3,111,294 adverse events.Moreover,we determined the organ toxicity scores of these drugs in 18 organ systems based on Medical Dictionary for Regulatory Activities(MedDRA).Of the 15 most toxic drugs,7 were either warned by the US FDA or withdrawn from the market in some countries and regions due to serious adverse events.This result indicates the new model performs well.Moreover,with the input of drug structure,we constructed organ toxicity prediction models for 18 selected organs in basis of Multilayer Perceptron(MLP)algorithm.The model performance was evaluated using ten-fold cross validation strategy.As the results,the organ models achieved an average coefficient of determination of 0.74.In particular,the hepatotoxicity prediction model was evaluated by comparing with the Drug-Induced Liver Injury Rank(DILIrank)dataset,which yielded a good AUC of 0.71.In Chapter Four,we took drug-induced glaucoma(DIG)as an example to explore the molecular mechanism underlying drug ocular toxicity.We conducted data mining of tremendous historical adverse drug events and genome-wide drug-regulated gene signatures to identify glaucoma-associated drugs.Upon these drugs,we carried out serial network analyses,including the weighted gene co-expression network analysis(WGCNA),to illustrate the gene interaction network underlying DIG.Furthermore,we applied pathogenic risk assessment to discover potential biomarker genes for DIG.As a result,we discovered 13 highly glaucoma-associated drugs,a glaucoma-related gene network,and 55 glaucoma-susceptible genes.These genes likely played central roles in triggering DIGs via an integrative mechanism of phototransduction dysfunction,intracellular calcium homeostasis disruption,and retinal ganglion cell death.Further pathogenic risk analysis manifested that a panel of nine genes,particularly OTOF gene,could serve as potential biomarkers for early-onset DIG prognosis.In summary,this thesis introduces an expert-free ADR severity evaluation algorithm for the first time.In basis of the ADR severity quantification,we develop a new machine learning model for broad-spectrum ADR profiling.As well,we build 18 models for organ toxicity assessment from simple input of drug structure.Similar work has been reported previously.Moreover,we unveiled possible mechanisms underlying drug-induced glaucoma.Nevertheless,these works provide not only new computational solutions for clinical drug organ toxicity assessment and molecular mechanism research,but also essential data and technical support for clinical safe chemotherapy and pharmacovigilance.Moreover,these works will help enhance the success rate of new drug discovery by reducing the attrition of drug safety problem.