Uncertainty-Aware Gene Rankings Reveal Key Players in Coexpression Networks

Motivation Key genes of a biological system are often prioritized by computing network science measures on a coexpression network inferred from transcriptomic data. But human population heterogeneity and modest sample sizes introduce uncertainty in the inferred coexpression network. Earlier studies have estimated this uncertainty using bootstrap resampling or similar approaches, but fewer have investigated how it propagates to downstream network analyses and affects gene prioritization. Methods and Results We present a systematic workflow to propagate network uncertainty to downstream measures such as degree/PageRank centrality, with the goal of producing robust gene scorings/rankings. We specifically propose uncertainty-aware scorings, BooNS and BPNS , which utilize the spread of a centrality measure across bootstrapped coexpression networks to prioritize stable central genes. Across several (semi-)simulated and real-world (GTEx) datasets, BooNS and BPNS recover reference or tissue-specific genes significantly better than other traditional centrality rankings. This performance gap highlights the long-overdue adoption of uncertainty-aware gene ranking for stable biological inference. Availability and Implementation Code and supplemental data are available at https://github.com/BIRDSgroup/Bootstrap-based_Node_Scorings_BNS , and https://tinyurl.com/BNS-suppl-data respectively.