Evaluation of the pKa's of Quinazoline Derivatives : Usage of Quantum Mechanical Based Descriptors

In this study, several quantum mechanical-based computational approaches have been used in order to propose accurate protocols for predicting the p<em>K_a</em>’s of quinazoline derivatives, which constitute a very important class of natural and synthetic compounds in organic, pharmaceutical, agricultural and medicinal chemistry areas. Linear relationships between the experimental p<em>K_a</em>’s and nine different DFT descriptors (atomic charge on nitrogen atoms (<em>Q</em>(N), ionization energy (<em>I</em>), electron affinity (<em>A</em>), chemical potential (m), hardness (h), electrophilicity index (w), fukui functions (<em>f ⁺</em>, <em>f ^-</em>), condensed dual descriptor (D<em>f</em>) and local hypersoftness (s⁽²⁾)) were considered. Several DFT methods (a combination of five DFT functionals and two basis sets) in conjunction with two different implicit solvent models were tested, and among them, M06L/6-311++G(d,p) level of theory employing the CPCM solvation model was found to give the strongest correlations between the DFT descriptors and the experimental p<em>K_a</em>’s of the quinazoline derivatives. The calculated atomic charge on N₁ atom (<em>Q</em>(N₁)) was shown to be the best descriptor to reproduce the experimental p<em>K_a</em>’s (R²=0.927), whereas strong correlations were also derived for <em>A</em>, w, m, and Δ<em>f</em>. In the last part, the applicability of isodesmic reaction scheme to the p<em>K_a</em> prediction of quinazoline derivatives was tested, and the calculated <em>A</em> was shown to be a well-established method for the classification of molecules, and thus, for the identification of a suitable reference molecule for the calculations. The QM-based protocols presented in this study will enable fast and accurate high-throughput p<em>K_a</em> predictions of quinazoline derivatives and the relationships derived can be effectively used in data generation for successful machine learning models for p<em>K_a</em> predictions.本研究采用多种基于量子力学的计算方法，旨在构建精准的预测方案，以测定喹唑啉衍生物的解离常数（pKₐ）——该类化合物是有机化学、药物化学、农业化学与医药化学领域中一类极为重要的天然及合成化合物类群。本研究考量了实验解离常数与九种不同密度泛函理论（DFT）描述符之间的线性关联，这些描述符包括：氮原子原子电荷（Q(N)）、电离能（I）、电子亲和能（A）、化学势（μ）、硬度（h）、亲电指数（ω）、福井函数（Fukui functions，f⁺、f⁻）、缩合双描述符（condensed dual descriptor，Df）以及局域超软度（local hypersoftness，s⁽²⁾）。本研究测试了多种DFT方法（五种DFT泛函与两种基组的组合）结合两种不同隐式溶剂模型的表现，结果发现，采用极化连续介质模型（CPCM）的M06L/6-311++G(d,p)理论水平，可实现DFT描述符与喹唑啉衍生物实验解离常数之间最强的相关性。研究表明，N₁原子的计算原子电荷（Q(N₁)）是复刻实验解离常数的最优描述符（决定系数R²=0.927），而电子亲和能A、亲电指数ω、化学势μ以及缩合双描述符Δf也均展现出较强的相关性。在研究的最后环节，本研究测试了等键反应方案（isodesmic reaction scheme）用于喹唑啉衍生物解离常数预测的适用性，结果证实，计算得到的电子亲和能A可作为分子分类的可靠方法，进而为计算筛选合适的参考分子提供依据。本研究提出的基于量子力学的计算方案，可实现喹唑啉衍生物快速且精准的高通量解离常数预测，所得关联关系亦可有效用于构建高性能机器学习模型的数据集生成，以实现解离常数的精准预测。