Home Contact FAQ

Jump to Matlab_code_help Python_code_help

Download manual in pdf

Other codes : using UniPath and chromatin interaction prediction after application of FITs

Using standalone executable for Linux

After downloading the file FITS.tar.gz gunzip and untar the file as such

$ gunzip FITS.tar.gz

$ tar -xvf FITS.tar

Now include the FITS folder in your PATH

$ export PATH=path_of_FITS_folder:$PATH

Now you can access commands of FITS from any folder.

For execution you have to pass filename of read-counts csv as a input file. Read-count csv file does not contain header and genomic location i.e. it consist of only data on which imputation is going to perform. Row represents sites and column represent samples/cells in csv file.

To get final imputed matrix, you run two phases1 of FITs using command

$ run_FITSPhase1.sh input=unimputed.csv output=imputed.csv

It is not necessary to use any option, however for faster executation you could use option

maxLevel= 2

maxLevel can be between 2 to 6 but not less than 2, such as

$ run_FITSPhase1.sh input=unimputed.csv maxLevel=2 output=imputed.csv

run phase1 of FITS many times, as such

$ run_FITSPhase1.sh input=unimputed.csv output=imputed.csv

$run_FITSPhase1.sh input=unimputed.csv output=imputed.csv

$run_FITSPhase1.sh input=unimputed.csv output=imputed.csv

....

$run_FITSPhase1.sh input=unimputed.csv output=imputed.csv

after executing Phase-1 many times, run the phase-2 of FITs

$run_FITSPhase2.sh input=unimputed.csv output=imputed.csv

For large read-count matrices

For imputing large read-count matrices without comsuming too much memory on computer use FITSPhase1L and FITSPhase2L

Such that you run FITSPhase1L many times ( more than 5 times) before you finally call FITSPhase2L

$ run_FITSPhase1L.sh input=unimputed.csv output=imputed.csv

$run_FITSPhase1L.sh input=unimputed.csv output=imputed.csv

$run_FITSPhase1L.sh input=unimputed.csv output=imputed.csv

……………………..

run_FITSPhase1L.sh input=unimputed.csv output=imputed.csv

Then run FITSPhase2L in same folder

$ run_FITSPhase1L.sh input =unimputed.csv output=imputed.csv

Using Matlab source code

"# FITS Matlab version"

You have to download matlab code in your local machine/server. For execution you have to pass filename of read-counts csv as a input file. Read-count csv file does not contain header and genomic location i.e. it consist of only data on which imputation is going to perform. Row represents sites and column represent samples/cells in csv file.

Start matlab and

Addpath of FITs using the command

>addpath(path_of_FITs_folder) ;

Now run phase1 of FITs on matlab console

> FITSPhase1 input='<csv file name>'

e.g.

> FITSPhase1 input='sce5_raw.csv'

'sce5_raw.csv' consist of epignome data corresponding to five cell type.

Other optional input parameter you can pass in phase1

>FITSPhase1 input='<csv file name>' output='<name to save imputed file>' maxLevel =<Depth upto which tree will grow>

By default maxLevel set to 4 and output set to 'FITSOutput'.

You can run FITSPhase1 parallely in background using

nohup matlab -nodisplay -nosplash -r "try FITSPhase1 input='<csv file name>'; catch; end; quit" > <name>.txt &

```

You can create n number of imputed matrix generated through phase1. Each run will generate imputed matrix.

Once Phase1 is over then run Phase2 to generate final imputed matrix based on matrix received as output from Phase1. You can run Phase2 on matlab console

> FITSPhase2 input='<csv file name>'

e.g.

> FITSPhase2 input='sce5_raw.csv'

You have to pass same input file as you passed in Phase1. Don't worry Phase2 takes only one minute to generate final output :)

Other optional input parameter you can pass in phase2

Ø FITSPhase2 input='<csv file name>' output='<name to save imputed file same as Phase1>' k =<topk correlated matrix feature/sample value use for final imputation> feature=<1/0 takes values either 1 or 0>

```

Default value for feature is zero. At value 0 phase2 will compute correlation among samples/cell (preffered) while value 1 will compute correlation features/sites wise.

For large read-count matrices

For imputing large read-count matrices without comsuming too much memory on computer use FITSPhase1L and FITSPhase2L

Such that you run FITSPhase1L many times ( atleast 5 times) before you finally call FITSPhase2L

Run FITSPhase1L many times before calling FITSPhase2L

FITSPhase1L input=unimputed.csv output=imputed.csv

Then run FITSPhase2L in same folder

$FITSPhase2L input=unimputed.csv output=imputed.csv

As FITSPhase1L occupies less RAM memory you could run multiple processes parralely using nohup

nohup matlab -nodisplay -nosplash FITSPhase1L input='<csv file name>’ output=’<imputed.csv>' ; > <name>.txt &

nohup matlab -nodisplay -nosplash -r FITSPhase1L input='<csv file name> ’ output=’<imputed.csv>' ; > <name>.txt &

.....

nohup matlab -nodisplay -nosplash -r "try FITSPhase1L input='<csv file name>’ output=’<imputed.csv>' ; catch; end; quit" > <name>.txt &

Then run FITSPhase2L

nohup matlab -nodisplay -nosplash -r "try FITSPhase2L input='<csv file name>’ output=’<imputed.csv>' ; catch; end; quit" > <name>.txt &

Using Python code

"# FITS Python version"

One needs to have python 3.0+ installed in their machine. You have to download python code in your local machine/server. For execution you have to pass filename of read-counts csv as a input file. Read-count csv file does not contain header and genomic location i.e. it consist of only data on which imputation is going to perform. Row represents sites and column reprent samples/cells in csv file.

$ python3 FITSPhase1.py -i <csv file name>

e.g.

$python3 FITSPhase1.py -i sce5_raw.csv

'sce5_raw.csv' consist of epignome data corresponding to five cell type.

Other optional input parameter you can pass in phase1

$ python3 FITSPhase1.py -i <csv file name> -o <name to save imputed file> -l <Depth upto which tree will grow>

Usage help can be availed by following command

$ python3 FITSPhase1.py -h

By default -l (maxLevel) set to 4 and -o (output) set to 'FITS_OUTPUT'.

You can run FITSPhase1 parallely in background using :

$ nohup python3 FITSPhase1.py -i <csv file name> > <name>.txt &

You can create n number of imputed matrix generated through phase1. Each run will generate imputed matrix. Once Phase1 is over then run Phase2 to generate final imputed matrix based on matrix received as output from Phase1.

$python3 FITSPhase2.py -i <csv file name>

e.g.

$python3 FITSPhase2.py -i sce5_raw.csv

You have to pass same input file as you passed in Phase1. Don't worry Phase2 takes only one minute to generate final output :)

Other optional input parameter you can pass in phase2

$python3 FITSPhase2.py -i <csv file name> -o <output> -t <topk > -c <1/0 takes values either 1 or 0>

Here topk represents number of top correlated vectors to us to build final matrix.

Default value for feature is zero. At value 0 phase2 will compute correlation among samples/cell (preffered) while value 1 will compute correlation features/sites wise.

Usage help can be availed by following command

$ python3 FITSPhase2.py -h

For large read-count matrices

For imputing large read-count matrices without consuming too much memory on computer use FITSPhase1L and FITSPhase2L

Such that you run FITSPhase1L many times (5 times) before you finally call FITSPhase2L

Run FITSPhase1L many times before calling FITSPhase2L

nohup python3 FITSPhase1.py -i <csv file name> > <name>.txt &

Then run FITSPhase2L in same folder

$python3 FITSPhase2L.py input -i unimputed.csv -o imputed.csv