O6-甲基鸟嘌呤-DNA甲基转移酶（MGMT）是一种重要的细胞机制，用于修复由鸟嘌呤O6-烷基化剂诱导的潜在细胞毒性DNA损伤，能够使细胞对某些癌症化疗药物高度耐药。为了克服MGMT介导的肿瘤耐药性，已设计和合成了大量潜在的MGMT不活化剂。我们利用基于[3H]-甲基化DNA的MGMT不活化测定法确定这些化合物对人重组MGMT的不活化能力，并计算出IC50值。利用370个化合物的结果，我们进行了定量构效关系（QSAR）建模，以确定化学结构与MGMT不活化能力之间的相关性。建模基于将排序后的pIC50值或化学结构或者随机分割。模型方程中共有九个分子描述符，其机理解释表明氮原子的状态、脂肪族初级氨基团、在拓扑距离3处的O-S存在、Al-O-Ar/Ar-O-Ar/R..O..R/R-O-C=X存在、电离势和氢键给体是导致不活化能力的主要因素。最终模型具有较高的内部稳健性、拟合度和预测能力（R2pr = 0.7474，Q2Fn = 0.7375-0.7437，CCCpr = 0.8530）。确定最佳分割模型后，我们基于整个化合物集建立了完整模型，使用相同的描述符组合。此外，我们还使用基于相似性的汇入技术提高模型的外部预测能力（R2pr = 0.7528，Q2Fn = 0.7387-0.7449，CCCpr = 0.8560）。利用“预测可靠性指标”工具检查了66个真实外部化合物的预测质量。总之，我们定义了与MGMT不活化相关的关键结构特征，从而可以设计出可能改善癌症治疗临床效果的MGMT不活化剂。

O6-methylguanine-DNA methyltransferase (MGMT) constitutes an important cellular mechanism for repairing potentially cytotoxic DNA damage induced by guanine O6-alkylating agents and can render cells highly resistant to certain cancer chemotherapeutic drugs. A wide variety of potential MGMT inactivators have been designed and synthesized for the purpose of overcoming MGMT-mediated tumor resistance. We determined the inactivation potency of these compounds against human recombinant MGMT using [3H]-methylated-DNA-based MGMT inactivation assays and calculated the IC50 values. Using the results of 370 compounds, we performed quantitative structure-activity relationship (QSAR) modeling to identify the correlation between the chemical structure and MGMT-inactivating ability. Modeling was based on subdividing the sorted pIC50 values or on chemical structures or was random. A total of nine molecular descriptors were presented in the model equation, in which the mechanistic interpretation indicated that the status of nitrogen atoms, aliphatic primary amino groups, the presence of O-S at topological distance 3, the presence of Al-O-Ar/Ar-O-Ar/R..O..R/R-O-C=X, the ionization potential and hydrogen bond donors are the main factors responsible for inactivation ability. The final model was of high internal robustness, goodness of fit and prediction ability (R2pr = 0.7474, Q2Fn = 0.7375-0.7437, CCCpr = 0.8530). After the best splitting model was decided, we established the full model based on the entire set of compounds using the same descriptor combination. We also used a similarity-based read-across technique to further improve the external predictive ability of the model (R2pr = 0.7528, Q2Fn = 0.7387-0.7449, CCCpr = 0.8560). The prediction quality of 66 true external compounds was checked using the "Prediction Reliability Indicator" tool. In summary, we defined key structural features associated with MGMT inactivation, thus allowing for the design of MGMT inactivators that might improve clinical outcomes in cancer treatment.