Check out the new paper here! https://www.nature.com/articles/s43588-022-00263-8
Xiaoqiao’s method assembles minimal lists of highly informative genes using an iterative SVM approach. Each iteration uses only a small numbers of cells and genes for computation. She uses a clever ‘active’ strategy to prioritize cells and genes that are most commonly misclassified, thus ensuring that selected genes improve classification on outliers. Because only a small number of cells and genes are used at each step, the process is very memory efficient, and allows her to compute over huge datasets (>1M cells) in a little over an hour (compared to days for other algorithms).