Genetic Algorithm-Based Optimization of In Silico prediction Models for Oral Absorption

近年来，组合化学和高通量筛选的引入大大加快了药物发现的速度。然而，在这些新技术中发现的热门化合物大多在药代动力学方面存在严重问题。因此，现在期望开发预测化合物的药代动力学性质的方法。本研究的目的是建立一种预测方法，用于预测药物的口服吸收，考虑到大多数药物是口服剂型。首先，我们开发了一种遗传算法结合偏最小二乘法探索最佳组合的分子预测描述符。简而言之，假设必要和不必要的描述符分别被描述为“1”和“0”，则描述符的组合被表示为二进制序列。利用遗传算法寻找具有最高可预测性的最优二进制序列。作为分子描述符，我们选择了拓扑指数，可以很容易地计算的基础上的二维分子结构。基于该方法，我们可以获得良好的预测模型的水溶性和上皮细胞通透性，这两个主要因素决定口服生物利用度

In recent years, the rate of drug discovery has been much accelerated by introduction of combinatorial chemistry and high-throughput screening. However, the hit compounds found in such new technologies mostly have serious problems in pharmacokinetics. Therefore, it is now being desired to develop the methods of predicting the pharmacokinetic properties of compounds. The purpose of this investigation was to establish a prediction method for predicting oral absorption of drugs, considering that a majority of drugs are in oral dosage forms. First, we developed a genetic algorithm-combined partial least squares method for exploring the optimal combination of molecular descriptors for prediction. In brief, assuming that necessary and unnecessary descriptors were depicted as "1" and "0", respectively, the combination of descriptors was expressed as a binary sequence. The optimal binary sequence, which possesses highest predictability, was explored by the use of genetic algorithm. As molecular descriptors, we selected topological indices that can be easily calculated based on 2-dimensinal molecular structure. Based on the approach, we could obtain good prediction models for aqueous solubility and epithelial cell permeability, that are two major factors determining oral bioavailability