Supervised dimensionality reduction for exploration of single-cell data by HSS-LDA

Data science through the lens of systems immunology

“…While our manuscript was biology focused, much of our work extends beyond the biology and into data science. 2 This makes Patterns a great fit for the scope of our manuscript.…”

“…After completing the project, we realized that HSS-LDA had considerable utility for a variety of biological applications and technologies, so we fully explored the topic in this new manuscript. 2 …”

“…DRG: While I launched the idea, my co-first author Meelad Amouzgar was the real driving force for the current paper. 2 Meelad rotated in Sean Bendall’s group as I was wrapping up my PhD. He came into the group as an already accomplished computational biologist and was very excited and motivated by the project.…”

“…I’m curious about the future of supervised dimensionality reduction in single-cell data beyond our manuscript. 2 …”

A supervised take on dimensionality reduction via hybrid subset selection

“… 6 Thus, semi-supervised or supervised dimensionality reduction approaches that use group labels may better capture biological structure corresponding to these labels of interest, rather than unsupervised approaches that attempt to preserve all components of variance in the data but end up suffering from large distortions. Motivated by some of these considerations, in this issue of Patterns , Bendall and colleagues propose a new method 7 —hybrid subset selection coupled to linear discriminant analyses (HSS-LDA)—to address some of these limitations ( Figure 1 ).

Figure 1 Conceptual comparison of the supervised HSS-LDA approach to existing unsupervised dimensionality reduction techniques

…”