Construction Of The Knowledge Bases For Drug Transporters And Metabolizing Enzymes And Discovery Of Drug Response Mechanisms

In the clinical treatment of many diseases,the response to the same dose of the same drug varies significantly from patient to patient,and the adverse drug reactions and drug failures that result have become an important factor in the health of patients.In recent years,with the proposal of precision medicine plans focusing on individual differences,a new light has been shone on the treatment of diseases by changing the traditional clinical drug use model.The aim is to make more accurate dosage and drug selection based on the individual characteristics of each patient,to achieve greater efficacy and fewer adverse effects,which is important for disease diagnosis,rational drug use,and reduction of clinical mortality.With the continuous refinement of various omics technologies,the rapid development of artificial intelligence methods,and advances in high-performance computing,significant achievements have now been achieved in the field of precision medicine.However,there are still significant challenges in transforming existing data and tools into guidance for clinical drug use practices.In particular,the incomplete data system related to important protein-related data that causes individual differences in drug reactions,insufficient tools for translating existing data into clinical practice,and incomplete understanding of the mechanisms of drug reactions significantly hinder the development and clinical application of precision medicine.Based on the challenges mentioned above,the following four research were carried out in this dissertation.First,an open-accessible database,the variability of drug transporter database(VARIDT)was established.Through a comprehensive literature review and collection,266 experimentally validated DTs were confirmed,for the first time,by the transporting drugs approved and in clinical/preclinical,providing an important reference indicator for clinical studies of drugs.Secondly,for each confirmed DT,VARIDT provided comprehensive information on multiple aspects of DT variability,including(i)epigenetic regulations and genetic polymorphisms of DTs;(ii)disease-,species-and tissue-specific protein abundances of DTs;and(iii)exogenous factors modulating the activity/expression of DTs,by systematically integrating existing literature data and developing and applying genomics data integration techniques.In addition,VARIDT enabled the first interplay analysis among multiple aspects of variability by building user-friendly website pages and adopting a more rational website infrastructure,which is of great significance in revealing the interactions between complex factors affecting drug transport,providing rigorous big data support for inspiring new therapeutic strategies and enabling AI-assisted precision medicine.The database can be accessed,queried,and downloaded for free via this link: https://idrblab.org/varidt/.Second,a newly developed database,interactome of drug-metabolizing enzyme(INTEDE)was constructed.Through a systematic literature review and collation,448 host DME and 599 microbial DME were confirmed,for the first time,using their metabolizing drugs,encompassing the most comprehensive experimentally validated DME data available and providing important reference information for drug metabolism studies.In addition,for each identified DME,its interactome network(including microbiome-DME interaction,host protein-DME interactions,and xenobiotics-DME interactions)was constructed via a comprehensive review of available literature resources,application of genomic/methylation omic data processing and integration technique,and user-friendly website visualization technique.More importantly,INTEDE allowed data to be downloaded and cross-analyzed between different types of DME interactions for free.This vast,connected,and structured data provided by INTEDE is expected to have a profound impact on the future of clinical treatment optimization and personalized medicine.The INTEDE database can be accessed for free via this link: https://idrblab.org/intede/.Third,a deep learning-based drug response prediction model,named DD-Response,was developed.DD-Response was the first to establish a novel two-dimensional mapping characterization method based on feature correlation,which re-characterizes high-dimensional and highly heterogeneous drug molecule features and cell line gene expression features into two-dimensional images,capturing the intrinsic correlation information between features.It addresses the over-fitting of models caused by high-dimensional features.Secondly,a multi-channel convolutional neural network relying on two pathways stacked with multiple layers respectively,incorporating a priori knowledge of the biology of gene pathways,efficiently learns drug features and cell lineage features.Finally,a three-layer fully connected neural network was used to learn drug-cell line interaction and achieve accurate prediction of drug-cell response.As the result,DD-Response reconstructed the intrinsic association between information to present not only the variability and intrinsic association of drug molecule profiles but also the variability of cancer types based on disease-related gene signature patterns.Furthermore,DD-Response demonstrated excellent predictive power on drug-cell line response prediction tasks,providing technical support to accelerate the achievement of precision medicine.Fourth,exploring the mechanisms that influence the balance of drug safety and efficacy for individualized drug use.By constructing a signature system of human PPI networks and biological system profiles to characterize targets of narrow therapeutic index(NTI)drugs in cancer and cardiovascular diseases,and applying artificial intelligence methods to explore for the first time the sophisticated NTI definitions between cancer and cardiovascular diseases and the underlying mechanisms behind them.Ultimately,10 common features and four unique features were identified that characterize NTI drug targets and NNTI drug targets for these two disease groups.Further analysis of the features revealed that particular attention should be paid to the ability of compound targets to be hubs and the efficiency of target signaling in the human PPI network during the development of compounds for these two diseases.Together,these newly identified features elucidated key factors influencing the NTI properties of anti-cancer and anti-cardiovascular disease drugs and may offer guidance in assessing the efficacy and safety of drug candidates and circumventing adverse clinical drug reactions.In conclusion,by constructing two databases related to drug transporters and drug metabolizing enzymes,which are important proteins causing individual differences in drug response,and based on the constructed high-quality data and publicly available datasets,this dissertation applied artificial intelligence methods to develop drug response prediction models and explored the mechanism of drug safety and efficacy balance,laying the foundation for big data-driven gene-directed drug use and helping to promote the early realization of precision medicine.