Abstract:

Carrier-free self-assembled nanomedicine has attracted wide attention and become one of the powerful strategies for cancer treatment due to its unique advantages such as simple preparation process, high drug loading, low cost and avoiding the toxicity caused by carriers. However, with millions of possible drug combinations, determining whether they can form self-assembling nanoparticles is a major challenge. With indomethacin as the model drug, different types of anti-tumor drugs were selected to explore their ability to form self-assembled nanoparticles with indomethacin, and different methods such as Hansen solubility parameters, machine learning model, binding energy calculation and COSMO-RS theory were utilized to study the self-assembly behavior. The self-assembly process was visualized by molecular dynamics simulation and quantitative chemistry calculation was used to analyze the molecular interaction to reveal the driving force of the self-assembly process. It was found that the machine learning model based on the training of high-throughput experimental values can quickly predict the probability of self-aggregation and co-aggregation with indomethacin molecules, and the combination of drug molecules can be preliminarily screened. In addition, through the study of thermodynamic mechanism, suitable drug molecule combinations can be selected from the perspective of energy and charge distribution, including the comparison of binding energy between the drug molecule itself and two different drug molecules, Hansen solubility parameter difference and surface charge density distribution. The self-assembly behavior is predicted using the Hansen solubility parameter model, COSMO-RS theory, and binding energy acquisition descriptors, and compared with the prediction of the machine learning model based on molecular fingerprints as descriptors. Based on molecular dynamics simulation, it was found that the self-assembly of drug molecules to form nanoparticles is a spontaneous aggregation behavior. Further analysis of the weak interaction between molecules revealed that hydrogen bond interaction is the key factor driving the self-assembly of drug molecules. Based on the research results of this paper, the different eigenvalues of drug molecules calculated by the thermodynamic model can be coupled into the machine learning model to enhance the physical meaning of the machine learning model, and an intelligent screening platform for self-assembled nanomedicine can be established, providing important guidance for the design and preparation of carrier-free nanomedicine delivery system with high delivery efficiency and combination therapy.

Key words: thermodynamic model, machine learning, quantum chemistry calculation, molecular simulation, self-assembling nanoparticles

摘要：

以吲哚美辛作为模型药物，选用不同种类的抗肿瘤药物探究其与吲哚美辛形成自组装纳米粒的能力，采用Hansen溶解度参数模型、COSMO-RS理论模型、结合能计算等不同手段获得描述符预测纳米自组装行为与基于分子指纹作为描述符的机器学习算法预测进行对比验证，并通过分子动力学模拟可视化分子自组装过程以及量化计算进行分子间相互作用能分析发现氢键为自组装过程关键驱动力。研究表明，机器学习算法可快速预测药物分子自聚集和与吲哚美辛分子共聚集的概率初步筛选出药物分子组合，进一步结合热力学理论模型研究比较药物分子自身与不同药物分子间结合能大小、Hansen溶解度参数差值以及表面电荷密度分布等描述符可更准确地筛选合适的药物组合，为设计制备递送效率高、可联合治疗的无载体纳米药物递送系统提供重要的理论指导。

关键词: 热力学模型, 机器学习, 量化计算, 分子模拟, 自组装纳米粒