微管靶向药物紫杉醇（PTX）和多西他赛（DTX）是广泛使用的化疗药物。然而，细胞凋亡过程、微管结合蛋白和多药耐药蛋白的失调可能会改变紫杉醇类药物的疗效。在这篇综述中，我们利用公开的药理学和全基因组分析数据集，构建了多CpG线性回归模型，预测了PTX和DTX药物的活性，这些数据集使用了数百个组织来源不同的癌细胞系。我们的研究显示，基于CpG甲基化水平的线性回归模型可以高精度地预测PTX和DTX的活性（相对于DMSO的生存率对数倍变化）。例如，287个CpG位点的模型可以在399个细胞系中以0.985的R2预测PTX的活性。同样精确的是342个CpG位点的模型，可以在390个细胞系中以0.996的R2预测DTX的活性。然而，与基于CpG的模型相比，我们的预测模型（采用mRNA表达和突变作为输入变量的组合）精度较低。虽然一种包含290个mRNA / 突变的模型能够以0.830的R2（针对546个细胞系）预测PTX的活性，一种包含236个mRNA / 突变的模型能够计算对553个细胞系的DTX活性，R2为0.751。基于CpG的模型限制在肺癌细胞系中也具有很高的预测能力（PTX R2≥0.980，74 CpG, 88 cell lines；DTX R2≥0.980，58 CpG, 83 cell lines）。这些模型反映了紫杉醇类药物活性/耐药的分子生物学机制。实际上，PTX或DTX CpG模型中的许多基因功能与细胞凋亡（如ACIN1、TP73、TNFRSF10B、DNASE1、DFFB、CREB1、BNIP3）和有丝分裂/微管相关（如MAD1L1、ANAPC2、EML4、PARP3、CCT6A、JAKMIP1）有关。还包括参与表观遗传调控的基因（HDAC4、DNMT3B以及组蛋白去甲基酶KDM4B、KDM4C、KDM2B和KDM7A）以及以前从未与紫杉醇类药物联系起来的基因（DIP2C、PTPRN2、TTC23、SHANK2）。总之，完全基于多个CpG位点的甲基化水平，可以准确预测细胞系中的紫杉醇类药物活性。版权所有© 2023 Elsevier Inc.保留所有权利。

The microtubule-targeting paclitaxel (PTX) and docetaxel (DTX) are widely used chemotherapeutic agents. However, the dysregulation of apoptotic processes, microtubule-binding proteins, and multi-drug resistance efflux and influx proteins can alter the efficacy of taxane drugs. In this review, we have created multi-CpG linear regression models to predict the activities of PTX and DTX drugs through the integration of publicly available pharmacological and genome-wide molecular profiling datasets generated using hundreds of cancer cell lines of diverse tissue of origin. Our findings indicate that linear regression models based on CpG methylation levels can predict PTX and DTX activities (log-fold change in viability relative to DMSO) with high precision. For example, a 287-CpG model predicts PTX activity at R2 of 0.985 among 399 cell lines. Just as precise (R2=0.996) is a 342-CpG model for predicting DTX activity in 390 cell lines. However, our predictive models, which employ a combination of mRNA expression and mutation as input variables, are less accurate compared to the CpG-based models. While a 290 mRNA/mutation model was able to predict PTX activity with R2 of 0.830 (for 546 cell lines), a 236 mRNA/mutation model could calculate DTX activity at R2 of 0.751 (for 531 cell lines). The CpG-based models restricted to lung cancer cell lines were also highly predictive (R2≥0.980) for PTX (74 CpGs, 88 cell lines) and DTX (58 CpGs, 83 cell lines). The underlying molecular biology behind taxane activity/resistance is evident in these models. Indeed, many of the genes represented in PTX or DTX CpG-based models have functionalities related to apoptosis (e.g., ACIN1, TP73, TNFRSF10B, DNASE1, DFFB, CREB1, BNIP3), and mitosis/microtubules (e.g., MAD1L1, ANAPC2, EML4, PARP3, CCT6A, JAKMIP1). Also represented are genes involved in epigenetic regulation (HDAC4, DNMT3B, and histone demethylases KDM4B, KDM4C, KDM2B, and KDM7A), and those that have never been previously linked to taxane activity (DIP2C, PTPRN2, TTC23, SHANK2). In summary, it is possible to accurately predict taxane activity in cell lines based entirely on methylation at multiple CpG sites.Copyright © 2023 Elsevier Inc. All rights reserved.