Assessing the impact of transcriptomics data analysis pipelines on downstream functional enrichment results

Abstract

Transcriptomics, and in particular RNA-Seq, has become a widely used approach to assess the molecular state of biological systems. To facilitate its analysis, many tools have been developed for different steps, such as filtering lowly expressed genes, normalisation, differential expression, and enrichment. While numerous studies have examined the impact of method choices on differential expression results, little attention has been paid to their effects on further downstream functional analysis using enrichment of gene sets, such as pathways, which typically provides the basis for interpretation and follow-up experiments. To address this gap, we introduce FLOP (FunctionaL Omics Processing), a comprehensive nextflow-based workflow that combines various methods for preprocessing and downstream enrichment analysis, allowing users to perform end-to-end analyses of count level transcriptomic data. We illustrate FLOP capabilities on diverse datasets comprising samples from end-stage heart failure patients and cancer cell lines in both basal and drug-perturbed states. We found that the correlation between gene set enrichment analysis results can vary significantly for alternative pipelines. Additionally, we observed that not filtering the data had the highest impact on the correlation between pipelines in the gene set space, especially in settings with limited statistical power. Overall, our results underscore the impact of carefully evaluating the consequences of the choice of preprocessing methods on downstream enrichment analyses. We envision FLOP as a valuable tool to measure the robustness of functional analyses, ultimately leading to more reliable and conclusive biological findings.

Robust Evaluation of Deep Learning-based Representation Methods for Survival and Gene Essentiality Prediction on Bulk RNA-seq Data

“…The RNA-Seq data was preprocessed following a classical bioinformatics pipeline. While different preprocessing options were tested (data and results not shown) as they can impact downstream analyses (Paton et al 2023), we decided to choose a fixed standard choice to keep a reasonable computational budget and compare models all else being equal. For all the experiments without pre-training on external datasets, we selected the 5,000 most variable genes on the training sets, applied a logarithmic operation and normalized the data with mean-standard scaling.…”