Cells:
LUHMES cells from 3 batches were merged together into 1 analysis. All cells have only a single type of gRNA readout. Quality control resulted in 8708 cells.
Genes:
Only genes detected in > 10% of cells were kept, resulted in 6213 genes.
Normalization:
Seurat "LogNormalize": log(count per 10K + 1).
Batch effect, unique UMI count, library size, and mitochondria percentage were all corrected for. The corrected and scaled expression data were used as input for subsequent factor analysis.
Here, our "guide", \(G\) matrix, consists of 15 types (14 genes + NTC) of gene-level knock-down conditions across cells.
Gibbs sampling was initiated from SVD, and conducted for 2000 iterations; the posterior mean estimates were averaged over the last 500 iterations.
Examples of factor ~ perturbation associations:
Number of genes that passed GSFA LFSR < 0.05 under each perturbation:
Overlap of GSFA DEGs between perturbations:
Venn diagram for DEGs found under perturbations ADNP, ARID1B, ASH1L, CHD2:
SFARI high risk (score 1, 2) genes among the DEGs:
Target genes: Genes w/ non-zero loadings in each factor (PIP cutoff at 0.95);
Backgroud genes: all 6213 genes included in factor analysis;
Statistical test: hypergeometric test (over-representation test);
Only GO terms/pathways within the size of 10 ~ 500 and have have an enrichment FC > 2 and FDR < 0.05 are kept.
Gene sets: Gene ontology "Biological Process" (non-redundant).
We used the "Wang" method in GOSemSim
to measure the similarity between GO BP terms, and all the significant terms in factors of interest were further grouped into 9 clusters using hierarchical clustering with the "ward.D" agglomeration method. The clustering results of all these GO BP terms are stored here.
Terms of interest
Factors of interest
Gene sets: The Reactome pathway database.
Factors of interest
PTEN is not detected in > 10% cells.
According to Figure 4E of the reference paper, among these marker genes, mature neuron marker genes are down-regulated and negative regulator genes are up-regulated in ADNP, CHD2 and ASH1L knock-downs; while the opposite happens in PTEN knock-down.
Here we show all the marker genes that passed LFSR < 0.05 under each knock-down. In general, their effect sizes are consistent with the findings of the paper.