Background: Although many prognosis-predicting molecular scores for breast cancer have been developed, they are applicable to only limited disease subtypes. We aimed to develop a novel prognostic score that is applicable to a wider range of breast cancer patients.
Methods: We initially examined The Cancer Genome Atlas breast cancer cohort to identify potential prognosis-related genes. We then performed a meta-analysis of 36 international breast cancer cohorts to validate such genes. We trained artificial intelligence models (random forest and neural network) to predict prognosis precisely, and we finally validated our prediction with the log-rank test.
Findings: We identified a comprehensive list of 184 prognosis-related genes, most of which have been not extensively studied to date. We then established a universal molecular prognostic score (mPS) that relies on the expression status of only 23 of these genes. The mPS system is almost universally applicable to breast cancer patients (log-rank P < 0.05) in a manner independent of platform (microarray or RNA sequencing).
Interpretation: The mPS system is simple and cost-effective to apply and yet is able to reveal previously unrecognized heterogeneity among patient subpopulations in a platform-independent manner. The combination of mPS and clinical stage stratifies prognosis even more precisely and should prove of value for avoidance of overtreatment. In addition, the prognosis-related genes uncovered in this study are potential drug targets. FUND: This work was supported by KAKENHI grants from the Ministry of Education, Culture, Sports, Science, and Technology of Japan to H.S. (19K20403) and to K.I·N (18H05215).
Keywords: AI; Breast cancer; Personalized medicine; Prognosis; Scoring system.
Copyright © 2019 The Authors. Published by Elsevier B.V. All rights reserved.