Identifying Genetic Markers: Dimension Reduction and Feature Selection for Sparse Data

To develop a principled suite of software on two methods for marker selection and one on data imputation or matrix completion in scRNA-seq data.

This group worked to develop solutions for the often inefficient clustering potential of single-cell technologies by adapting a 1-bit compressed sensing algorithm to allow for the better separation of cell clusters using only a small number of genes. To complement the compressed sensing approach, they also developed a mutual information framework to build markers out of a more extensive set of statistically significant genes to maximize clustering ability while minimizing redundancy between markers.