Crucially, these second order features require high measurement sensitivity as ligands and receptors in neighboring cells need to be co-detected. Downsampling analysis showed the ability to detect predictive LR pairs drops with loss of sensitivity, but not imaging area. (6/7)

We found that LR+ cell pairs have a different gene expression program and these programs can predict glioma recurrence times in bulk RNAseq data. Thus, Cell-signaling specific gene expression programs rather than cell-type specific gene expression programs may be important. (5/7)

We found that second-order features (i.e pairwise interactions) at the DNA and RNA level are highly predictive of whether tumors will MT or reamin stable. These are either chromosomes structure alternations or ligand-receptor signaling across neighboring cells. (3/7)

We applied high sensitivity DNA and RNA seqFISH (~50% sensitivity) combined with a Cell Interaction Foundation model on biopsies from 18 IDH-mutant glioma patients who either malignant transformed (MT) at later date or remained stable. (2/7)

Quality of the data matters! at least for predicting glioma progression from a patient's initial biopsy before any progression has occurred. Excited to share our new pre-print, authored by Michal Polonsky, Jonathan Fox, Yunrui Lu and Sheel Shah. (1/7) www.biorxiv.org/content/10.6...