Helmholz Heike, Resuli Redon, Tacke Marius, Nourisa Jalil, Tomforde Sven, Aydin Roland, Willumeit-Römer Regine, Zeller-Plumhoff Berit
Institute of Metallic Biomaterials, Helmholz-Zentrum Hereon, Geesthacht, Germany.
Institute of Material Systems Modeling, Helmholz-Zentrum Hereon, Geesthacht, Germany.
Comput Struct Biotechnol J. 2025 Jun 8;27:2711-2718. doi: 10.1016/j.csbj.2025.06.023. eCollection 2025.
Angiogenesis is one of the first stages in fracture healing and bone repair. Therefore, numerous studies evaluating the effect of Mg as a promising degradable, metallic biomaterial on the proliferation and function of endothelial cells have been performed. However, these studies lack methodological homogeneity and therefore differ in fundamental conclusions. Here, Mg-concentration-, donor- and cell age- dependent relations to primary human umbilical cord vein endothelial cells (HUVEC) proliferation and migration were investigated systematically. The generated data were utilized to develop regression models in order to assess and predict the cell response on Mg exposition in a concentration range of 2-20 mM Mg in cell culture medium extract. A concentration of > 2 mM already induced a detrimental effect in the sensitive primary HUVECs. Molecular data quantifying angiogenesis markers supported this finding. An increased migration capacity has been observed at a concentration of 10 mM Mg. We compared linear regression, random forests, support vector machines, neural networks and large language models for the prediction of HUVEC proliferation for a number of scenarios. Using these machine learning methods, we were able to predict the proliferation of HUVECs for missing Mg concentrations and for missing passages with mean absolute errors below 10 % and as low as 8.5 %, respectively. Due to strong differences between the cell behaviour of different donors, information for missing donors can be predicted with mean absolute errors of 15.7 % only. Support vector machines with linear kernel performed best on the tested data, but large language models also showed promising results.
血管生成是骨折愈合和骨修复的最初阶段之一。因此，已经进行了大量研究，评估镁作为一种有前景的可降解金属生物材料对内皮细胞增殖和功能的影响。然而，这些研究缺乏方法学上的同质性，因此在基本结论上存在差异。在此，系统地研究了镁浓度、供体和细胞年龄与原代人脐静脉内皮细胞（HUVEC）增殖和迁移之间的关系。利用生成的数据建立回归模型，以评估和预测在细胞培养基提取物中镁浓度为2-20 mM范围内暴露时细胞的反应。浓度>2 mM就已经对敏感的原代HUVEC产生了有害影响。量化血管生成标志物的分子数据支持了这一发现。在镁浓度为10 mM时观察到迁移能力增强。我们比较了线性回归、随机森林、支持向量机、神经网络和大语言模型在多种情况下预测HUVEC增殖的能力。使用这些机器学习方法，我们能够分别以低于10%和低至8.5%的平均绝对误差预测缺失镁浓度和缺失传代次数时HUVEC的增殖情况。由于不同供体的细胞行为存在很大差异，对于缺失供体的信息，只能以15.7%的平均绝对误差进行预测。线性核支持向量机在测试数据上表现最佳，但大语言模型也显示出了有前景的结果。
