适用于细针穿刺样本的DNA甲基化乳腺癌诊断模型的构建与验证OA

OBJECTIVE:To develop a DNA methylation-based diagnostic model for breast cancer and to evaluate its diagnostic performance in breast fine-needle aspiration(FNA)cytological specimens,thereby laying a foundation for development of an economically feasible molecular testing tool to aid in the diagnosis of FNA specimens.METHODS:This retrospective study utilized 98 formalin-fixed paraffin-embedded(FFPE)breast tissue specimens as the training cohort and 209 breast FNA cytological specimens as the test cohort.Targeted methylation sequencing using next-generation sequencing was performed.The methylation haplotype load(MHL)values of individual methylation loci in FFPE specimens were subjected to weighted analysis,and the loci with weight in the top 20 were selected as methylation biomarkers.Bayesian,random forest,and support vector machine(SVM)algorithms were applied to construct a diagnostic model for breast cancer.Histopathological diagnosis was used as the gold standard to evaluate diagnostic performance of the methylation model.RESULTS:In the FFPE specimens,20 methylation loci including SOX17,CALN1,KDM4B,SND1 and LINC01622,which had the largest weights in MHL,were selected to establish breast cancer diagnostic models.In the FNA testing cohort,the random forest based model achieved an area under the receiver operating characteristic curve(AUC)of 0.890[95%CI(0.833,0.948)],with a sensitivity of 100.00%[95%CI(97.25%,100.00%);136/136],a specificity of 78.08%[95%CI(67.32%,86.03%);57/73],an accuracy of 92.34%[95%CI(87.93%,95.23%);193/209],a positive predictive value of 89.47%[95%CI(83.59%,93.42%);136/152],and a negative predictive value of 100.00%[95%CI(93.69%,100.00%);57/57].The diagnostic performance of the random forest based model was significantly superior to that of the Bayesian(AUC=0.781)and SVM(AUC=0.740)based models(all P<0.05).CONCLUSION:A DNA methylation-based diagnostic model incorporating methylation features of genes such as SOX17,CALN1,KDM4B,SND1 and LINC01622,combined with a random forest algorithm,demonstrated excellent diagnostic performance in breast FNA specimens.