Methodology Development Of In Silico ADMET Prediction And Drug Design Targeting Estrogen Receptors

This thesis concentrated on some important problems existing in computer aided drug design, including two parts:(1) methodology development of in silico ADMET prediction; and (2) drug design targeting estrogen receptors. In the first chapter, a brief introduction about the research background and basic concepts was given, including genetic algorithms, support vector machine, molecular fingerprint, etc. Emphases were put on those concepts and principles tightly related to this thesis, such as molecular dynamics simulation, molecular docking and scoring. Finally, we discussed the current situation of drug design research in China.ADMET is one of the most important issues in drug discovery pipeline. With technical improvement, people realized that it was not difficult to find potent and specific leads in the early stage of drug discovery. However, experimental evaluation of ADMET properties can still not meet the demands of lead discovery and optimization due to the time-and cost-effectiveness. Therefore, in silico ADMET prediction has become a practicable alternative choice so far, which could break through the "bottleneck" in the high throughput drug discovery process. The first part of this thesis is developing new methodology for in silico ADMET prediction. The second chapter described a genetic algorithm based variable selection method. In this method, an improved fitness function involving with the cross validation coefficients was used, and we also designed an elite warehouse to avoid the pitfall of conventional genetic algorithms. Then the logBB models were built as the case study. Molecular descriptors were widely used in conventional ADMET prediction methodologies. However, there are several shortcomings of using molecular descriptors. The third chapter described a chemical classification method based on substructure recognition. Therefore the molecular descriptors were avoided during the model building. In addition, information gain analysis was involved to evaluate each substructure of the molecules, which could help to interpret the machine learning models from medicinal chemistry perspective. We also presented their applications in the ADMET prediction.Estrogen receptors (ERs) belong to nuclear receptor superfamily, which play crucial roles in human body, including the growth and differentiation of reproductive system, central nervous system and skeletal system. They are also important drug target for some diseases, including breast cancer. At present, most clinic use drugs are selective ER modulator. Although the underlying mechanism remains unclear, they have indeed alleviated the side effect of drugs. The second subtype, ERβ, which was discovered in 1996, shedding light on discovering subtype selective ligands for achieving tissue selective drugs. Therefore, many recent works focused on the differences in biological functions and pathways between the two subtypes. Besides, finding novel selective ligands could not only benefit the drug development, but also provide the chemical sensor for biological researches. The second section of this thesis is focusing on discovering selective ERβligands with various molecular modeling and CADD methods. In chapter 4, we performed steered molecular dynamics simulation to investigate the dynamic process of ligand unbinding from two different subtypes of ERs. In this work, we firstly find out that the process of ligand unbinding also contribute to the ligand selectivity. Accordingly, we proposed two suggestions for improving ERP selectivity. In chapter 5, we discovered 18 potent ligands of ERβwith virtual screening based on a structure optimized through molecular dynamics simulation. Among them, dual profile was observed in two ligands, as agonists for ERβand antagonists for ERa, which might be served as lead compounds for selective ER modulators. The results also suggest that structures optimized through MD are applicable to lead discovery. Besides, the structure activity relationship results confirmed the suggestions proposed in chapter 4. In these work, we presented an integrated work which involved theoretic calculations and experiments. The results have been verified for each other.