Abstract 7P
Background
The RNA modification have been correlated with patient’s survival in various cancers. However, whether RNA modification genes are associated with molecular classification and potential application in breast cancer patients’ prognostication and prediction of disease-free survival (DFS) is largely unknown. Here we aim to develop and validate of a deep learning RNA modification model to predict DFS in patients with breast cancer, and further to analysis the association of the RNA modification model with immune microenvironment.
Methods
In this study, a total of 2,236 non-metastatic breast cancer patients were included. We built the DeepSurv in TensorFlow, a deep-learning survival neural network-based model on 2,149 non-metastatic breast cancer patients' data with 112 RNA modification genes in training cohort from The Cancer Genome Atlas (TCGA) and NCBI Gene Expression Omnibus (GEO). The algorithm was externally validated on an independent test cohort, comprising 87 non-metastatic breast cancer patients in Sun Yat-sen Memorial Hospital of Sun Yat-sen University. The primary end point was DFS.
Results
In the training cohort, we established a deep-learning survival neural network model showed promising results to predict DFS risk score in breast cancer patients; patients with low-risk vs high-risk score had statistically significantly longer DFS [hazard ratio (HR) = 14.88, 95% CI = 11.96–18.53, P < 0.005]. The RNA modification risk score was associated with DFS (AUCs for 2-, 3-, 5-, and 7-year survival were 0.92, 0.93, 0.93 and 0.94, respectively) superior molecular subtype, tumor, node, and metastasis staging system. Similarly, in the test cohort, patients with low-risk vs high-risk score had statistically significantly longer DFS (P < 0.005), the AUCs for 2- and 3-year survival were 0.89, and 0.89, respectively. The RNA modification model was validated across various subgroups. Further, this study identified that model obviously related to multiple tumor immune microenvironment characteristics.
Conclusions
This study developed and validated novel deep learning survival neural network model showed reliable individual DFS information in prognostic evaluation and treatment recommendation in patients with breast cancer.
Legal entity responsible for the study
Beijing Normal University-Hong Kong Baptist University United International College, Hong Kong Baptist University.
Funding
Has not received any funding.
Disclosure
All authors have declared no conflicts of interest.