Characterizing gene expression responses to biomechanical strain in anin vitromodel of osteoarthritis

Abstract

Osteoarthritis (OA) is a common chronic degenerative joint disease affecting articular cartilage and underlying bone. Both genetic and environmental factors appear to contribute to the development of this disease. Specifically, pathological levels of biomechanical stress on joints play a notable role in disease initiation and progression. Population-level gene expression studies of cartilage cells experiencing biomechanical stress may uncover gene-by-environment interactions relevant to OA and human joint health. To build a foundation for such studies, we applied differentiation protocols to develop an in vitro system of chondrogenic cell lines (iPSC-chondrocytes). We characterized gene regulatory responses of three human iPSC-chondrocyte lines to cyclic tensile strain treatment. We measured the contribution of biological and technical factors to gene expression variation in this system and, even in this small sample, found several genes that exhibit inter-individual expression differences in response to mechanical strain, including genes previously implicated in OA. Expanding this system to include iPSC-chondrocytes from a larger number of individuals will allow us to characterize and better understand gene-by-environment interactions related to OA and joint health.

Non-negative matrix factorization algorithms greatly improve topic model fits

“…We run these comparisons on four data sets (Table 1): two text data sets [28,53] that have been used to evaluate topic modeling methods (e.g., [3,64]); and two data sets from single-cell RNA sequencing (scRNA-seq) experiments [51,67]. While the second application may be less familiar to readers, recent papers have illustrated the potential for topic models to uncover structure from scRNA-seq data, in particular structure that is not well captured by (hard) clustering [18,66,5,40]. See Appendix A.…”

“…Finally, we used two measures to assess the quality of the computed solutions to (7): the change in the loss function (L, F) or, equivalently, the change in the Poisson NMF log-likelihood, log p PNMF (X | L, F); and the maximum residual of the KKT conditions (39,40).…”