I believe the parent refers to this [1,2,3] study. Indeed, this was about targeting many (11,923) genes with Perturb-seq (CRISPR screen with single-cell RNA-sequencing readout). There are two human cell lines used in the study (K562 and RPE1). For functional annotation, authors focused on 1,973 targeted genes that had strong transcriptional phenotype after the perturbation. As there's some correlation structure, that's what they studied, annotating clusters of individual perturbations using public databases (like STRING [4]) and literature. Seems like a lot of great work has been done here though stating that we now know all the functions of all the genes might be a bit of a stretch indeed.