FACTORs: a Novel R Package for Functional Data Reduction and Hypothesis Testing

Abstract

ABSTRACTTo extract biological meaning from transcriptomics analysis, investigators almost exclusively rely on biological annotation databases and the subsequent associations made between gene products and ontology terms (i.e., gene ontology analysis). Although there is ease and utility to this approach, multiple hypothesis testing methods such as differential expression analysis are performed in the absence ofin silicovalidation, and downstream gene ontology analysis methods lack precision and propagate Type I error. Therefore, we present a novel R package, Functional Association Vectors (FACTORs), that begins to address some of the common pitfalls associated with functional association studies in transcriptomics research. FACTORs are vectorized containers for directly comparing and testing congruent functional association statistics at the molecular level. Our goal was to develop novel methodology an R software package to allow for the experimental validation of differentially expressed genes that are conserved across studies and to reduce Type I error in the down-stream functional analysis of these signals. FACTORs are generalizable and flexible, allowing for any association statistic such as log fold change (logFC), t-statistic, p-value, or other functional significance score. To demonstrate utility of FACTORs in the cross-study validation of global mRNA expression profiling, we used differential expression analysis summary statistics obtained from two studies with publicly available transcriptomic data. Through this demonstration we show FACTORs provide a more precise and generalizable functional hypothesis testing methodology and data reduction approach that directly tests functional association statistics at the molecular level, across experiments.AUTHOR SUMMARYWe present a novel statistical methodology and software tool that can be used to validate transcriptomic data across studies. FACTORs is available as an R package and will aid in transcriptomic data reduction, identifying gene expression profiles that are conserved across studies, and improving the precision and generalizability of functional association studies. We propose FACTORs as a generalizable methodology for reducing Type I error and increasing the biological relevance of functional associations in transcriptomics studies.