The significance of molecular heterogeneity in breast cancer batch correction and dataset integration

Breast cancer research benefits from a substantial collection of gene expression datasets that are commonly integrated for subsequent computational analysis. Gene expression batch effects arising between experimental batches, where technical signal differences confound true biological variation, must be addressed when integrating datasets. Several approaches exist to address these batch effect differences. This brief report clearly demonstrates that popular batch correction techniques can significantly distort vital biomarker expression signals. Through the implementation of batch correction and visualisation of integrated expression values, we profile the extent of these distortions and evaluate different variations of Combat batch correction on key breast cancer biomarker expression values. The diversity of breast cancer as a molecularly heterogenous disease is well recognised. The potential impact of this heterogeneity on vital dataset processing and downstream evaluation remains under-evaluated. We believe this short study presents the first analysis of the interplay between dataset molecular composition and concomitant robustness of integrated, batch-corrected expression signal of breast cancer biomarker expression.