Researchers in the Departments of Statistics and Data Science and Mathematics, Florentina Bunea, Marten Wegkamp, and their Ph.D. student, Xin Bing, developed a new scalable method for determining significant, interpretable, latent factors that can explain outcomes of interest in complex biological data. Grounded in theoretical work, the power of the method lies in its capability to identify a simpler, but hidden, structure of high dimensional biological input. The identification process allows for unambiguous labeling of the latent mechanism. This advances the current state-of-the-art paradigms centered around spectral methods, which cannot directly identify latent factors. An added, specialized, step allows for downstream inference based on the revealed hidden mechanisms.
Building upon two previous articles, in partnership with Jishnu Das’s lab at the University of Pittsburgh, the Cornell researchers further refined this method for direct applications across biological domains. The approach has been used to uncover drivers of a range of molecular, cellular and organismal phenotypes. Using single-cell RNA sequencing (scRNA-seq) data, they were able to identify known and novel transcriptomic signatures reflecting cell-intrinsic and cell-extrinsic circuits (across myeloid cells, fibroblasts and keratinocytes) underlying the pathophysiology of diffuse systemic sclerosis. Using spatial transcriptomic data, SLIDE was able to capture immune cell partitioning by spatial localization in lymph nodes in an allergic asthma model. Using matched scRNA-seq and T cell receptor sequencing data, SLIDE also identified drivers of clonal expansion of CD4 T cells in a mouse model of type 1 diabetes.
This effort has been recognized by the scientific community. It appeared in Nature Methods online on February 19th, 2024, and will appear in print in the May 2024 issue. This work was also highlighted by Nature Methods in a Research Briefing.
