Code associated with tafka-crisprQTL (Gasperini, et al. 2019)
The following scripts/files were used in differential expression testing. We also include information where to access relevant data files in the Data section below.
custom_monocle_da_functions.R: other scripts source this very slightly modified version of monocle's differential expression testing functions (they report the intercept and the coefficient for the target term in the model). As such, this file must remain in the same directory as the scripts.
The scripts used for differential expression testing in the pilot and at-scale experiments are very similar but we keep them distinct for clarity:
-
get_deg.pilot.R: differential expression script used for the pilot experiment -
get_deg.at_scale.R: differential expression script used for the at-scale experiment
Each of these two scripts has the same interface, for example::
Rscript get_deg.pilot.R pilot_highmoi_screen.cds.rds \
grna_group \
gene_gRNAgroup_pair_table.pilot.txt \
output_file.txt
Note that for the at-scale experiment, in order to cut down on computational costs, we used a random subset of 50,000 cells from the experiment as the reference set rather than including all cells. This means that all cells with a guide to the specified target are included, but only cells in the 50,000 cell reference are included if they do not have a guide. This means that the call to this script is slightly different:
Rscript get_deg.at_scale.R at_scale_screen.cds.rds \
grna_group \
gene_gRNAgroup_pair_table.at_scale.txt \
output_file.txt \
--reference_cells_rds 50k_reference_cells.rds
These scripts run the relevant DEG tests for a single target at a time (corresponding to grna_group). We then parallelized these over all grna_groups via our cluster. A list of all ggrna_group values for the pilot and at-scale screens are provided in the Data section below as a reference.
Available GEO GSE120861.
at_scale_screen.cds.rds: Processed scRNAseq CellDataSet for the at-scale screen (with metadata such as gRNAs assigned to each cell) to input intoget_deg.at_scale.R.pilot_highmoi_screen.cds.rds: Processed scRNAseq CellDataSet for the pilot high MOI screen to input intoget_deg.pilot.R.grna_groups.pilot.txt: gRNAgroup identifiers for the pilot screen and their corresponding spacer sequences. The first column is a comprehensive list of all values to be used in place ofgrna_groupin theget_deg.pilot.Rcommand above. Only one group should be passed to this script at a time.grna_groups.at_scale.txt: gRNAgroup identifiers for the at-scale screen at corresponding spacer sequences. The first column is a comprehensive list of all values to be used in place ofgrna_groupin theget_deg.at_scale.Rcommand above. Only one group should be passed to this script at a time.50k_reference_cells.rds: File storing the names of the 50,000 cells used used for differential expression testing for the at-scale screen (passed as--reference_cells_rdstoget_deg.at_scale.R).gene_gRNAgroup_pair_table.pilot.txt: Table of pilot screen's gene-target (gRNAgroup) pairs considered for differential expression testing (with corresponding information columns about each target and candidate target gene).gene_gRNAgroup_pair_table.at_scale.txt: Table of at-scale screen's gene-target (gRNAgroup) pairs considered for differential expression testing (with corresponding information columns about each target and candidate target gene).all_deg_results.pilot.txt: All differential expression results from the pilot screen, along with 'empirical p-values*' (only calculated forgRNA_groupsassociated with decreases in candidate target gene's expression).all_deg_results.at_scale: All differential expression results from the at-scale screen, along with 'empirical p-values*' (only calculated forgRNA_groupsassociated with decreases in candidate target gene's expression).zero_inflated_outlier_genes_to_exclude.atscale.txt: We observed that genes with a very high mean expression value relative to what you would expect given the number of cells in which they are expressed tended to show up frequently as false positives in permutation tests. These may be outliers with respect to their dispersion given that monocle uses shrinkage to estimate smoothed dispersion values when parameterizing the regression model used for differential gene expression. To avoid this source of false positives, we ignored this small set of outliers in downstream analysis. This set was derived from the data stored inat_scale_screen.cds.rds.zero_inflated_outlier_genes_to_exclude.pilot.txt: Same as above, but a list of outlier genes we identified from thepilot_highmoi_screen.cds.rds.
* An empirical P-value was defined for each gene-gRNAgroup pair test that decreased the candidate target gene's expression. The empirical P-value was defined as: [(the number of NTCs with a smaller P-value than that test’s raw P-value) + 1] divided by [the total number of NTCs tests + 1].
-
Gasperini2019.at_scale_screen.cand_enhancer_x_exprsd_genes.200503.csv: The gene x candidate enhancers interactions we used to call hits from the 'at-scale' experiment. There are 78,562 interactions here, but in the manuscript we state 78,776 were tested - this latter sum total refers to unique sites targeted across both the pilot and 'at-scale' experiments. In the "at scale" experiment from which we drew hits, there were 78,562 unique enhancer-x-gene interactions. Additionally, we state we targeted 5,779 in candidate enhancers in the 'at-scale' experiment: in hindsight, this should have been stated as 5,723 (56 candidate enhancers we targeted did not have any genes above the cell % expression threshold in the surrounding 2Mb region. Please see manuscript for details re: expression threshold). Only 5,723 are included in this file. -
at_scale_screen.phenoData.txt.gz: Monocle pData object used to track gRNA-cell associations. The column names (in order) and their associated explanations:
sample: sample ID
cell: each cell's identifier, as assigned by cell ranger
total_umis: total UMIs assigned to the cell
Size_Factor: metric typically used by Monocle to account for variation in total UMI counts across cells
gRNAgroups: A gRNAgroup is defined as all the gRNAs that are targeting the same candidate enhancer or positive control site (usually 2 gRNAs designed per group, see Gasperini et al methods for further explanation). If a gRNAgroup was detected in this cell by our analysis, its name is present in this string. The GEO file that contains the gRNAgroup names and their associated sequences for the at-scale screen is: GSE120861_grna_groups.at_scale.txt.gz.
gRNAgroups_dups: If both the gRNAs in a gRNAgroup are present in the same cell, the gRNAgroup's name is present twice in this column (a kind of silly column).
gRNAsequences: If a gRNAsequence is detected in this cell by our analysis, its name is present in this string.
gRNA_read_count: number of reads associated with the gRNAs
gRNA_umi_count: UMIs associated w the gRNAs
[the remaining columns were output by a tool we use, but probably are not generally useful:]
gRNAproportion: proportion of total reads that match gRNAs in this list - note we do not suggest using this column as it appears
guide_count: number of gRNAs detected in the cell by our analysis
sample_directory: duplicate of the sample ID column
bc_file: guide barcode file ID
batch_ID1: overall prep batch
batch_ID2: reagent lot prep batch
batch_ID3: within batch chip ID
batch_ID4: within chip lane ID
mito: percentage mitochondrial
To parse the gRNA-cell associations using this file, you can use either of two columns: gRNAsequences or gRNAgroups. Here's a helpful line that could help you parse the gRNA-cell association columns.
pData(gene_cds)$gRNA_detected <- grepl(paste0('(^|_)', gRNA_group, '(_|$)'), GSE120861_at_scale_screen.phenoData.df$gRNAgroups_or_gRNAsequences_column) #this outputs a true/false input if a gRNA has been detected
If you would like to convert this into a matrix representation, you could do something like the following -- go through each unique gRNAgroup in the gRNAgroup-sequence file (or each guide sequence, as found in GSE120861_grna_groups.at_scale.txt.gz) and ask which cells are annotated as having a guide belonging to the group (or the guide itself) using the regex above. This would give you a vector corresponding to the 0/1 status for that group across all cells. If you do this for all groups you would be able to construct the full cell by group (or gRNA sequence) matrix.
The following packages are required to run the scripts:
library(argparse)
library(dplyr)
library(Matrix)
library(monocle)
library(stringr)
library(tidyr)
Note that we used Monocle 2. Because 1) monocle differential expression testing has not changed for some time and 2) we load our own differential expression functions that are modified to report intercepts and the target coefficient, monocle versioning should not have practical implications as long as the objects stored in our provided RDS files load without error.