Lies, Damn Lies, and Genomics

Genomic tools have revolutionized our ability to investigate biological questions at unprecedented resolution, but with this power comes a responsibility to critically assess the data, methods, and inferences we produce. In this talk, I explore how assumptions, batch effects, and analytic flexibility can lead to erroneous conclusions even in high-profile publications, and how such pitfalls are often hidden beneath layers of statistical confidence and computational output. By highlighting real examples from ecological genomics, I show how different aligners, priors, and analytical choices can drastically alter evolutionary interpretations, particularly those involving selection inference, trait association, and comparative genomics. From replicate failures in model organisms to challenges in inferring soft sweeps and gene function, I argue that many of the strongest claims in the literature rest on methodological ground that is less stable than it appears. Using case studies ranging from butterflies to sticklebacks to large-scale genome consortium datasets, I emphasize the importance of cross-validation, independent replication, and reanalysis as essential tools in genomic science. I also reflect on my own lab’s work, where seemingly solid genotype-to-phenotype associations revealed unexpected layers of complexity when examined more closely. In sum, this talk offers a critical, and ultimately hopeful, reality check for genomic researchers. Our goal is not to distrust data, but to better understand its limitations and to refine our questions, designs, and interpretations in ways that make us better scientists.