Development of mechanism specific subgroup classification method for breast cancer recurrence prediction

Abstract

The most important clinical challenge in the treatment of breast cancer is the prevention of recurrence in early breast cancer patients. Receptor status-based patient stratification, gene expression-based recurrence risk pre-diction, and additional chemotherapy, which are currently being used as standard treatment in clinical practice, are only available in some patient groups and cannot even explain the heterogeneity of patient responses. This study hypothesized that the heterogeneity of these patient responses is due to the mechanism-specific sub-groups present in breast cancer patients, and developed a methodology to detect recurrence by specifically recognizing these mechanism-specific subgroups. The study consists of three parts

construction of a breast cancer-specific marker database (BRCA-KB), construction of a breast cancer gene expression database (BRCA-GEX), and construction of SynBI-Marker, a pipeline to optimize disease mechanism subgroup-specific classifi-cation model. BRCA-KB is a marker-mechanism association database for each breast cancer mechanism that was not provided in the existing marker databases. As a result of evaluating the marker-mechanism relationship built in the database through gene recovery tests using the marker prioritization algorithm, each gene showed high support in terms of similarity between biological networks and functional groups. Based on the relation-ship, a new marker-mechanism association was inferred and verified through literature survey, confirming that the relationship between the marker and the mechanism in the context of cancer is supported. BRCA-GEX is the database that provides the most comprehensive gene expression of the breast cancer databases proposed to date. It features pre-processing using a consistent pipeline and reconfirmation of patient clinical information through literature survey. Since the gene expression data related to breast cancer recurrence is provided in the database for each cohort, it is possible to discover and verify breast cancer recurrence markers in various condi-tions. The SynBI-Marker is a pipeline for optimizing the maximum probability ensemble model of specific discriminant markers for subgroups of disease mechanisms. It consists of four parts: prioritization of markers, optimization of marker sets, final ensemble model construction and disease mechanism annotation. By apply-ing the pipeline to the problem of predicting breast cancer recurrence in a specific cohort provided by BRCA-GEX in the mechanism marker feature space derived from BRCA-KB, we were able to obtain superior discrimi-nant performance compared to existing breast cancer recurrence markers. In conclusion, the BRCA-KB, BRCA-GEX, and SynBI-Marker Pipeline built through this study were able to discover specific patient groups for breast cancer recurrence mechanisms and to establish a mechanism identification model with better classifica-tion performance. This system can be used as a methodology to derive markers of good performance with better interpretability and can be extended to other diseases.