Speaker: Li-Xuan Qin, Memorial Sloan Kettering Cancer Center, "Evidence-Based Practice of Omics Data Harmonization for Survival Outcome Prediction"

Abstract: Survival analysis plays an important role in biomedical transcriptomics studies for deriving reliable predictors of patient prognosis and treatment response. While survival analysis methods are available to address the issues of high dimensionality and signal sparsity, reproducible translation of transcriptomics data to survival predictors has been hampered by the ubiquitous presence of data artifacts associated with disparate experimental handling. Published studies often deal with these artifacts by borrowing data harmonization methods (such as quantile normalization and ComBat) devised for differential expression analysis, based on unfounded optimism rather than supportive empirical evidence. We first illustrate important caveats of employing these methods for survival prediction, both intuitively and numerically using re-sampling-based simulations. We then develop a new harmonization method, dubbed “BatMan”, and assess its performance compared with ComBat and quantile normalization. We further demonstrate the advantage of BatMan utilizing microRNA data for ovarian cancer from the Cancer Genome Atlas. This work is in collaboration with Andy Ni (Ohio State University) and Mengling Liu (New York University).