SPLASH GLM-based supervised test

The repository contains the script for performing multinomial regression with lasso to identify and visualize anchors with metadata-dependent target fraction

Inputs:

• directory: SPLASH run directory
• metadata_file: path to metadata file
• run_name: prefix for the output folder
• datatype: can be either 10x when input data is 10x or non10x when input data is not 10x
• input_anchors: list of specific anchors
• sample_fraction_cutoff: cutoff for the minimum fraction of samples each anchor should be present
• num_anchors: maximum number of anchors to be tested

Two options for providing anchors:

• providing a list of anchors: each anchor in the list will be considered for the GLM test
• specifying sample_fraction_cutoff and num_anchors: In this case, among the anchors in result.after_correction.scores.csv that are in at least sample_fraction_cutoff of samples, and have avg_hamming_distance_max_target >5, the top num_anchors with the highest effect size are selected to be tested by the GLM.

example commands:

• specific list of anchors:

Rscript SPLASH_spervised_anchor.R /oak/stanford/groups/horence/Roozbeh/SPLASH_10x/runs/CCLE_all/ /oak/stanford/groups/horence/Roozbeh/SPLASH_10x/utility_files/CCLE_metadata_modified.tsv one_unaligned_target_anchors /oak/stanford/groups/horence/Roozbeh/SPLASH_10x/runs/CCLE_all/concatentaion_based_classified_compactors_one_SJ_one_unaligned.tsv

• anchors with the highest effect size:

Rscript SPLASH_spervised_anchor.R /oak/stanford/groups/horence/Roozbeh/SPLASH_10x/runs/CCLE_all/ /oak/stanford/groups/horence/Roozbeh/SPLASH_10x/utility_files/CCLE_metadata_modified.tsv one_unaligned_target_anchors 0.4 20000

metadata file format:

non-10x data:

The only requirement for metadata file is that the first column and second columns must be for sample names and the assigned class/group/category/celltype to each sample. Also, the first row must be column names (column names can be anything). The two columns are subsequently renamed as "sample_name" and "group" in the script. Below is an example metadata file:

sample_name	type
SRR8788980	carcinoma
SRR8788981	melanoma
SRR8788982	lymphoma
SRR8788983	carcinoma
SRR8657060	carcinoma

10x data:

Meta data file for 10x data must have an extra column (second column) for cell barcodes. Other requirements are the same as non-10x data:

sample_name cell cell_type
S2	AAACCCAAGACACACG	AT2
S1	AAACCCAAGCCATGCC	NK
S1	AAACCCAAGGGAACAA	Mac
S4	AAACCCAAGGGTTGCA	NK

Output files:

The output directory for the GLM test is: directory/run_name_supervised_metadata/.

• GLM_supervised_anchors.tsv: the output file containing anchors with non-zero GLM_coefficients whose largest GLM coefficient is greater than 1.
• plots: this subfolder contains the plots for visualizing the anchors with metadata-dependent target fraction varaition. For each anchor, two plots will be generated in a pdf named as anchor sequence, a box plot for showing the fraction of target1 per sample grouped by the category and a scatterplot for showing the counts for target1 and target2 per sample colorcoded by the metadata category.

Contact

Please contact Roozbeh Dehghannasiri ([email protected]).