Julia Salzman

The Salzman lab views genomic data as a statistical signal processing problem, leveraging tools from statistics, traditional machine learning and deep learning for biological discovery. We have recently introduced a new unifying paradigm, SPLASH (Statistically Primary aLignment Agnostic Sequence Homing), a data-compressive signal extraction that bypasses the historical computational machinery of genomics. Ongoing work extends the scope of this approach, making many problems in genomic sequence analysis easily accessible to computational engineers and statisticians. We are using SPLASH and its sister algorithms to train biological deep learning models at unprecedented scale, and leveraging these models to predict phenotypes and from single cell analyses of human and non-model organisms to drug responses in microbes, and other phenotypes across the tree of life, including in the study of plants, viruses, oceanic systems and their symbioses. The reference-free nature of SPLASH and its sister approaches enable dually the prediction of biological behavior – eg phenotype and attribution of the features responsible for these predictions.