Kalyakulina Alena, Yusipov Igor, Trukhanov Arseniy, Franceschi Claudio, Moskalev Alexey, Ivanchenko Mikhail
Artificial Intelligence Research Center, Institute of Information Technologies, Mathematics and Mechanics, Lobachevsky State University, Nizhny Novgorod 603022, Russia.
Institute of Biogerontology, Lobachevsky State University, Nizhny Novgorod 603022, Russia.
Int J Mol Sci. 2025 Jun 29;26(13):6284. doi: 10.3390/ijms26136284.
We present EpInflammAge, an explainable deep learning tool that integrates epigenetic and inflammatory markers to create a highly accurate, disease-sensitive biological age predictor. This novel approach bridges two key hallmarks of aging-epigenetic alterations and immunosenescence. First, epigenetic and inflammatory data from the same participants was used for AI models predicting levels of 24 cytokines from blood DNA methylation. Second, open-source epigenetic data (25 thousand samples) was used for generating synthetic inflammatory biomarkers and training an age estimation model. Using state-of-the-art deep neural networks optimized for tabular data analysis, EpInflammAge achieves competitive performance metrics against 34 epigenetic clock models, including an overall mean absolute error of 7 years and a Pearson correlation coefficient of 0.85 in healthy controls, while demonstrating robust sensitivity across multiple disease categories. Explainable AI revealed the contribution of each feature to the age prediction. The sensitivity to multiple diseases due to combining inflammatory and epigenetic profiles is promising for both research and clinical applications. EpInflammAge is released as an easy-to-use web tool that generates the age estimates and levels of inflammatory parameters for methylation data, with the detailed report on the contribution of input variables to the model output for each sample.
我们展示了EpInflammAge，这是一种可解释的深度学习工具，它整合了表观遗传和炎症标志物，以创建一个高度准确、对疾病敏感的生物年龄预测器。这种新方法弥合了衰老的两个关键特征——表观遗传改变和免疫衰老。首先，来自同一参与者的表观遗传和炎症数据被用于人工智能模型，该模型从血液DNA甲基化预测24种细胞因子的水平。其次，开源表观遗传数据（2.5万个样本）被用于生成合成炎症生物标志物并训练年龄估计模型。使用针对表格数据分析优化的最先进深度神经网络，EpInflammAge与34种表观遗传时钟模型相比，实现了具有竞争力的性能指标，在健康对照中总体平均绝对误差为7岁，皮尔逊相关系数为0.85，同时在多种疾病类别中表现出强大的敏感性。可解释人工智能揭示了每个特征对年龄预测的贡献。由于结合了炎症和表观遗传特征而对多种疾病具有敏感性，这在研究和临床应用方面都很有前景。EpInflammAge作为一个易于使用的网络工具发布，它可以为甲基化数据生成年龄估计值和炎症参数水平，并提供每个样本输入变量对模型输出贡献的详细报告。
