Comparison of the Metastasis Predictive Potential of mRNA and Long Non-Coding RNA Profiling in Systemically Untreated Breast Cancer

Our results suggested that classification followed by voting improved prognostic power using long non-coding RNAs compared to mRNAs regardless of classification strategy

Thi T. N. Do; Ines Block; Mark Burton; Kristina P. Sørensen; Martin J. Larsen; Martin Bak; Søren Cold; Mads Thomassen; Qihua Tan; Torben A. Kruse

2021

Scholarcy highlights

  • In breast cancer, clinical inter-tumor heterogeneity is represented by staging systems, whereas histopathologic and molecular classification reflect morphologic and genetic inter-tumor heterogeneity
  • Classification was conducted with a threshold that provided at least 90% sensitivity while maximizing specificity and performances assessed by leave-one-pair-out cross-validation using linear discriminant analysis, support vector machines based on a radial kernel or linear kernel, random forest, naïve
  • The difference between long non-coding RNAs and mRNA performance was significant at a p-value of 0.013
  • Classification was conducted with optimized balanced accuracy and performances assessed by leave-one-pair-out cross-validation using linear discriminant analysis, support vector machines based on a radial kernel or linear kernel, random forest, naïve Bayes, COX risk score, and logistic regression. a Mean of sensitivity and specificity
  • Classification with a fixed sensitivity of ≥90% for the individual machine learning methods followed by voting with a final sensitivity of 90%, obtained consistently higher overall accuracies when based on lncRNAs compared to mRNAs
  • An overall classification accuracy for mRNA data ranged from 53% using LR to 66% using RF
  • Classification was conducted with a threshold that provided at least 90% sensitivity while maximizing specificity and performances assessed by leave-one-pair-out cross-validation using linear discriminant analysis, support vector machines based on a radial kernel or linear kernel, random forest, naïve measures are indicated in percent. b p-value determined by a one-sided two-proportion z-test comparing the estimated significance between the seven machine learning methods using mRNA or lncRNA
  • Similar findings were observed using the alternative strategy of bAcc optimization and classification followed by voting suggested improved prognostic power using long non-coding RNAs compared to mRNAs

Need more features? Save interactive summary cards to your Scholarcy Library.