diff --git a/Snakefile b/Snakefile index b63c3fc..1f5afbf 100644 --- a/Snakefile +++ b/Snakefile @@ -1,5 +1,5 @@ import os -import fireducks.pandas as pd +import pandas as pd import splicekit DEFAULT_CORES = config["defaults"]["cores"] diff --git a/splicekit/core/annotation.py b/splicekit/core/annotation.py index 68aac6f..9416d10 100644 --- a/splicekit/core/annotation.py +++ b/splicekit/core/annotation.py @@ -134,23 +134,7 @@ def make_comparisons(): annotation.samples = [str(el) for el in annotation.samples] def write_comparisons(): - if splicekit.config.platform == 'SLURM': - job_rmats=""" -#!/bin/bash -#SBATCH --job-name={job_name} # Job name -#SBATCH --ntasks=1 # Number of tasks -#SBATCH --mem=8G # Allocate memory -#SBATCH --nodes=1 # All tasks on one node -#SBATCH --partition=short # Select queue -#SBATCH --output=logs/rmats/{comparison_name}.out # Output file -#SBATCH --error=logs/rmats/{comparison_name}.err # Error file - -{container} run_rmats --b1 results/rmats/{comparison_name}_test.tab --b2 results/rmats/{comparison_name}_control.tab --gtf {gtf_path} -t paired --readLength 150 --variable-read-length --allow-clipping --nthread 4 --od results/rmats/{comparison_name}_results --tmp results/rmats/{comparison_name}_temp - """ - - else: - - job_rmats=""" + job_rmats=""" #!/bin/bash #BSUB -J {job_name} # job name #BSUB -n 1 # number of tasks @@ -213,23 +197,7 @@ def write_comparisons(): write_mds_job() def write_edgeR_jobs(): - if splicekit.config.platform == 'SLURM': - job_edgeR=""" -#!/bin/bash -#SBATCH --job-name={job_name} # Job name -#SBATCH --ntasks=1 # Number of tasks -#SBATCH --nodes=1 # All tasks on one node -#SBATCH --mem=8G # Allocate memory -#SBATCH --partition=short # Select queue -#SBATCH --output=logs/edgeR/{atype}/{comparison_name}.out # Output file -#SBATCH --error=logs/edgeR/{atype}/{comparison_name}.err # Error file - -module load R -{container} R --no-save --args {input_folder} {atype} {control_name} {test_name} {comparison_name} {sample_membership} {filter_low} < {core_path}/comps_edgeR.R - """ - - else: - job_edgeR=""" + job_edgeR=""" #!/bin/bash #BSUB -J {job_name} # job name #BSUB -n 1 # number of tasks @@ -332,23 +300,7 @@ def write_edgeR_jobs(): fsh_acceptor_anchors.close() def write_dexseq_jobs(): - if splicekit.config.platform == 'SLURM': - job_dexseq=""" -#!/bin/bash -#SBATCH --job-name={job_name} # Job name -#SBATCH --ntasks=1 # Number of tasks -#SBATCH --nodes=1 # All tasks on one node -#SBATCH --mem=8G # Allocate memory -#SBATCH --partition=short # Select queue -#SBATCH --output=logs/dexseq/{atype}/{comparison_name}.out # Output file -#SBATCH --error=logs/dexseq/{atype}/{comparison_name}.err # Error file - -module load R -{container} R --no-save --args {input_folder} {atype} {control_name} {test_name} {comparison_name} {sample_membership} {samples} < {core_path}/comps_dexseq.R - """ - - else: - job_dexseq=""" + job_dexseq=""" #!/bin/bash #BSUB -J {job_name} # job name #BSUB -n 1 # number of tasks @@ -392,24 +344,7 @@ def write_dexseq_jobs(): fsh_exons.close() def write_mds_job(): - if splicekit.config.platform == 'SLURM': - job_mds=""" -#!/bin/bash -#SBATCH --job-name=edgeR_mds # Job name -#SBATCH --ntasks=1 # Number of tasks -#SBATCH --nodes=1 # All tasks on one node -#SBATCH --mem=8G # Allocate memory -#SBATCH --partition=short # Select queue -#SBATCH --output=logs/edgeR/mds/mds.out # Output file -#SBATCH --error=logs/edgeR/mds/mds.err # Error file - -module load R -{container} R --no-save --args {input_folder} {sample_membership} {filter_low} < {core_path}/comps_edgeR_mds.R - """ - - else: - - job_mds=""" + job_mds=""" #!/bin/bash #BSUB -J mds # job name #BSUB -n 1 # number of tasks diff --git a/splicekit/judge/__init__.py b/splicekit/judge/__init__.py index 42bf890..28af624 100644 --- a/splicekit/judge/__init__.py +++ b/splicekit/judge/__init__.py @@ -2,7 +2,7 @@ import sys import seaborn as sns import matplotlib.pyplot as plt -import fireducks.pandas as pd +import pandas as pd from scipy.stats.stats import pearsonr import glob import math