基于基质辅助激光解吸电离飞行时间质谱技术建立结直肠癌诊断模型及初步验证

Establishment and preliminary verification of diagnostic model for colorectal cancer based on matrix-assisted laser desorption ionization time-of-flight mass spectrometry technology

目的 基于基质辅助激光解吸电离飞行时间质谱(MALDI-TOF MS)技术建立结直肠癌(CRC)诊断模型,寻找CRC潜在的血清学标志物。方法 收集自2018年12月至2020年12月于联勤保障部队960医院就诊的114例CRC初诊患者(CRC组)和进行体检的60例健康体检者(健康组)的血清标本,按照3∶1的比例随机分为训练组131例(CRC患者87例,健康者44例)与验证组43例(CRC患者27例,健康者16例),分别用于模型的建立和初步验证。应用弱阳离子交换磁珠(MB-WCX)进行血清中低丰度蛋白提取纯化,经MALDI-TOF MS筛选CRC组和健康组中差异蛋白峰;基于3种算法(遗传算法、监督神经网络算法和快速分类算法)建立诊断模型,选取差异最显著的两个蛋白峰做聚类分析,将验证组数据带入诊断模型验证其敏感性、特异性及诊断效率。结果 CRC组与健康组蛋白指纹图谱存在明显差异,共筛选出9个有统计学意义的差异蛋白峰(曲线下面积>0.70),在CRC组表达上调7个,下调2个;其中,m/z 4645.32和m/z 5906.48差异最显著(P<0.01),曲线下面积分别为0.91、0.76;m/z 4645.32在CRC组表达显著下调,m/z 5906.48表达上调。比较发现监督神经网络模型诊断效能最佳,其敏感性为92.60%、特异性为81.25%、准确性为88.37%。结论 本研究建立的CRC诊断模型具有较好的诊断效能,其中,蛋白峰m/z 4645.32和m/z 5906.48有望成为CRC潜在的血清学标志物。

Objective To establish a diagnostic model for colorectal cancer(CRC)based on matrix-assisted laser desorption ionization time-of-flight mass spectrometry(MALDI-TOF MS),and to search for potential serum markers of CRC.Methods Serum samples were collected from 114 newly diagnosed CRC patients(CRC group)and 60 healthy persons who underwent physical examination(healthy group)in 960th Hospital of the PLA from December 2018 to December 2020.The patients were randomly divided into a training group of 131(87 CRC patients and 44 healthy individuals)and a validation group of 43(27 CRC patients and 16 healthy individuals)according to a 3∶1 ratio for model building and initial validation,respectively.The low abundance protein in serum was extracted and purified by magnetic bead-weak cation exchange(MB-WCX),and the differential protein peaks in CRC group and healthy group were screened by MALDI-TOF MS.The diagnostic model was established based on three algorithms(genetic algorithm,supervised neural network algorithm and fast classification algorithm),and the two protein peaks with the most significant differences were selected for cluster analysis,and the data of the verification group was brought into the diagnostic model to verify its sensitivity,specificity and diagnostic efficiency.Results There were significant differences in protein fingerprints between CRC group and healthy group.A total of 9 different protein peaks with statistical significance were screened(area under the curve>0.70),7 expressions were up-regulated and 2 expressions were down-regulated in CRC group.m/z 4645.32 and m/z 5906.48 showed the most significant difference(P<0.01),and area under the curve values were 0.91 and 0.76,respectively.The expression of m/z 4645.32 was significantly down-regulated in CRC group,while the expression of m/z 5906.48 was up-regulated.Comparison showed that supervised neural network model had the best diagnostic efficacy,with sensitivity of 92.60%,specificity of 81.25%and accuracy of 88.37%.Conclusion The CRC diagnostic model established in this study has good diagnostic efficacy,among which the protein peaks m/z 4645.32 and m/z 5906.48 are expected to be potential serum markers of CRC.