New and notable: Revisiting the "two cultures" through extrinsic noise

Abstract

In a classic article (1), Leo Breiman bears witness to the divergence between “two cultures” of statistics that emerged in the wake of readily accessible computing technology: the data modeling culture, which concerns itself with developing and fitting stochastic models, and the algorithmic modeling culture, which concerns itself with improving predictive accuracy without delving into unknown (and perhaps unknowable) mechanisms. More than two decades later, the distinct cultures of statistics are evident in approaches to single-molecule transcriptomics. The biophysics subfield focuses on assays that target a small number of genes and develops increasingly sophisticated mechanistic models, whereas the sequence census subfield uses descriptive, data-scientific methods such as those championed by Breiman.

Additional details