SpliceNet

To systematically explore the functions of core splicing factors and regulators in alternative splicing, RNA-seq analyses were carried out upon the knockdown of over 304 splicing-related factors.

Clone this repository on your computer

using CL:

git clone https://github.com/estepi/SpliceNet

Using RStudio:

• File -> New project -> Version Control -> GiT -> Clone a project from Git repository

Main scripts

Scripts associated to transcriptome-wide analysis of the effects of systematic knock down of splicing factors and regulators using siRNAs in HeLa cells.

• All scripts were written in R version >4.0

• PSI calculation was performed using VAST-TOOLS version

Workflow

Clean raw-data/ Replace NAs: Analyze NAs in VAST-TOOLS INCLUSION final Table:

• We count how many NAs contain a quality score for every PSI value computed. If they have more than 3, we replace this PSI value with NAnew3

• Input is the default vast-tools table (INCLUSION-...)

• Output: desired output, script 1: only replacing NAnew3

• If INCLUSION table was produced using last version of vast-tools, we can replace Intron Retention PSI value with NAI if the adjusted pvalue for the "read umbalance" is below 0.05.

• Input is the default vast-tools table (INCLUSION-...)

• Output: desired output, script 1: replacing NAnew3+NAI

How to run:

• 1.- Download scripts in your desired folder.
• 2.- Load libraries:

library(data.table)
library(stringr)

• 3.- Source the function you want to use:

source("replaceNA.R") # only NewNAs
source("replaceNAI.R") # NewNAs + NAIs

• Define the input / output files:

INCFile<-"test.tab"
OUTFile<-"test_NA_NewN3.tab"
OUTFile2<-"test_NA_NewN3_NAIs.tab"
replaceNA(INCFile, OUTFile)
replaceNAI(INCFile, OUTFile2)

Prepare tables for specific events

• script: prepareEVENTTable.R
• Requiered library: MatrixGenerics
• Input files (See files.md for more details)
  • PSI_full_No_Nas.txt
  • class_colors2020.txt
• Outputs:
  • dPSI values,
  • single (by columns, KDs) and double (by row and column) scaled deltapsi values

Network FDR computation

• This script computes Network FDR represented as the ratio betwenn TRUE links and RANDOM links. Input file is a matrix with EVENTS in rows and KDs (samples) in the column
• dPSI values can be none, single or double scaled (should be prepared in advance, see Prepare table section)
• The function retunrs the total number of links which are below a given correlation value and are significant (corrected pvalue <0.1).
• Expected values are higher FDR at lower correlation values
• Matrix randomization is prepared randomizing rows, so correlation between columns are disrupted. As noiser is the data, more random links will be find, higher FDR

• script: Net_fdr_CL_cor.R (use by CL or interactively)
• Requiered libraries: optparse, parallel, Hmisc, tibble, tidyr, utils, dplyr
• Parameters:

• -s 0.1 (start: from which correlation value the function scan the data. Correlation values are between 0 and 1)
• -e 0.4 (end: till which correlation value the funcion scan the data. Correlation values are between 0 and 1)
• -i 0.02 (interval: size of the interval to compute the correlation. For example, 0.02 means you will scan 0, 0.02, 0.04, 0.06, etc)
• -r 100 (number of random matrixes. Nuber of links in random data come from the mean of the number of links of those matrixs)
• -b (bin: folder where functions_fdr_MC_cor.R is)
• -c 12 (cores: number of cores to use)
• -f sscaled.tab (file: input file, should be prepared in advance. It contains only dPSI values, scaled or not ex: sscaled.tab)
• -n A3short (name: prefix to use for output files, ej: A3short)

• Usage using CL:

$R CMD Rscript ../SpliceNet/Net-fdr_CL_cor.R -f all_sscaled.tab -s 0.1 -e 0.2 -i 0.05 -r 5 -n short -b ../SpliceNet

• Output: The function returns a table and their corresponding plots for the number of links for real and random data in the interval of correlations FDR is computed as Number of Links RANDOM data / Number of links in REAL data * 100

For our files (see 'files.md'):

 ../SpliceNet/Net-fdr_CL_cor.R -s 0.1 -e 0.8 -i 0.01 -r 1000 -b ../SpliceNet/ -c 12 -f  A3_all_sscaled.tab -n A3long
 ../SpliceNet/Net-fdr_CL_cor.R -s 0.1 -e 0.8 -i 0.01 -r 1000 -b ../SpliceNet/ -c 12  -f A5_all_sscaled.tab -n A5long
 ../SpliceNet/Net-fdr_CL_cor.R -s 0.1 -e 0.8 -i 0.01 -r 1000 -b ../SpliceNet/ -c 12  -f ES_all_sscaled.tab -n ESlong
 ../SpliceNet/Net-fdr_CL_cor.R -s 0.1 -e 0.8 -i 0.01 -r 1000 -b ../SpliceNet/ -c 12  -f IR_all_sscaled.tab -n IRlong

Single Cor

Compute single correlation for a given dPSI table

• Script: singlecor.R(use by CL or interactively)

• Requiered libraries: optparse, utils, igraph, scales, Hmisc, tibble, tidy, utils, dplyr

• Input: Numeric matrix (only dPSI values, scaled or not), mininum correlation value, scripts folder, sample name

• Usage:

• Output: for a given threshold of correlation, it returns the edgelist:

• so, tg: source and target original order
• Pearson cor and absCor, with pvalue and fdr,
• source and target alphabetically ordered,
• link's name

• For our files:

 R CMD Rscript  ../SpliceNet/singlecor.R -m 0 -f all_sscaled.tab  -p  1 -n all -b ../SpliceNet

Extract centrality

For a given input matrix, it extracts degree (and normalized degree) poe each KD in a given interval of correlations It can help to identify high/low stable factors.

• Net-centrality_CL_cor.R (for command line)
• Net-centrality_interactive_cor.R (to run interactively)

generate_subset_from_GC_distribution.R