This study was only made possible with the incredible support of the Prostate Cancer Foundation @pcf-science.bsky.social

I love this tool and recommend it to anyone working with single cell data!

This work is the first publication from my lab and I could not be more proud!!!!
/end

We provide lots more details about this in our pre-print and helpful examples on how to run this for yourself in your own data using our package (sanin-lab.github.io/OARscRNA/).
7/n

We also use this method to predict how Dictyostelium discoideum changes their mitochondrial biology to go from an amoeba to multi-cellular aggregates! (Mitochondrial function sensing dye on the right lights up when cells aggregate and mitochondria lower membrane potential).
6/n

For example, in these plasmacytoid DCs, the transcriptional shift we detect is because of type I IFN production!
5/n

Cells where missing data patterns are not good predictors, score HIGHER because they have transcriptional programs that are different and therefore you should go and study why!
We are borrowing from statistical methods that examine missingness: doi.org/10.1007/s135...
4/n

2️⃣ We examine how well missing data patterns explain gene expression and derive a score from this result.
3/n

An OAR (Observed At Random) score links co-expression patterns and gene expression. A high OAR score indicates a poor prediction (an interesting cell!).
Our method has 2 main steps:
1️⃣ We define a set of co-expression patterns based on zero-counts, which we call missing data patterns.
2/n