APAIQ use bedGraph files as input, it can either take two bedGraph files from the forward strand and reverse strand, or single bedGraph file without strand information
The bedGraph file could be generated during the alighment using STAR with the option '--outWigType bedGraph --outWigNorm RPM'
The bedGraph could be also obtained from bam file by using genomeCoverageBed from BEDTOOLS
genomeCoverageBed -ibam file.bam -split -bg -scale scaled.factor
To get the value of reads per million (RPM), instead of raw count, a scaled factor should be specified.
The factor is equal to 1000000/(total unique mapped reads), and it should be caution to use '-du' if the data is paired-end.
To install the compiled version of APAIQ from conda, we recommend to creat a conda enviroment firstly using\
conda create -n apaiq_env python=3.7 and then
conda activate apaiq_env
conda install -c joshuachou apaiq.
All the enviroment files and required libraries would be installed automatically
To run APAIQ using source code from Github, please create a enviroment using the provided env files.
conda create --name apaiq_env --file apaiq.env.txt
To install each dependency package manually
conda create --name apaiq_env python=3.7
conda activate apaiq_env
conda install -c bioconda pybedtools
pip install tensorflow
conda install -c anaconda biopython
Fast install dependency manually, works for python 3.7 to 3.10
conda create --name apaiq_env
conda activate apaiq_env
pip install pybedtools
pip install tensorflow
pip install biopython
pip install pandas
To run with GPU, additional dependency should be installed by using the below code
conda install -c conda-forge tensorflow-gpu
to replace
pip install tensorflow
apaiq --input_file=RNAseq.depth.bedGraph --out_dir=out_dir/ --fa_file=genome_fa --name=sample_id --DB_file polyA.bed --model $model
a test data, pre-trained model and annotation db_file could be found through the link below:
https://drive.google.com/drive/folders/1KNj-dsh5hCmKI3dyhIsi_OuU6-3mLpBW?usp=sharing
Run APAIQ with source code:
python src_v2/APAIQ.v.1.2.py --input_file=RNAseq.depth.bedGraph --out_dir=out_dir/ --fa_file=genome_fa --name=sample_id --DB_file polyA.bed --model $model --t 30
Please use the prefix of the model files as the input of --model. For instance for the provided model in Google Drive, the option
should be --model snu398_model.ckpt
Run APAIQ for quantification
python regression/evaluateRegression.v.2.py --model=regression/regression.ckpt --factor_path=normalize_factor --pas_file=$pas_file --input_file=$input_file --out=$out_file --threshold=0
Please use the provided 'model', 'normalization factor' files. The pas_file should be in BED format and input_file is the same as the input_file used ing the scirpt for PAS identification.
The options for running APAIQ for PAS identification
--input_file <bedGraph file> input bedGraph file from strandless data
--input_plus <bedGraph file> input bedGraph file from forward strand
--input_minus <bedGraph file> input bedGraph file from reverse strand
--fa_file fasta file of the genome
--name sample name
--model model
--depth default=1. For unnormalized input. use --depth=10 for 10 millions total single-end mapped reads, 5 millions total paired-end mapped reads
--out_dir default='out_dir'. output directory.
--RNASeqRCThreshold default=0.05. Minimum RPM threshold for scaning item
--window default=201. Window sizes of the scaning item. Please do not change this value if you used the pre-trained model
--keep_temp use --keep_temp='yes', if you want to keep the temporary files.
The options for running regression model for PAS quantification
--input_file <bedGraph file> input bedGraph file from strandless data
--input_plus <bedGraph file> input bedGraph file from forward strand
--input_minus <bedGraph file> input bedGraph file from reverse strand
--pas_file file of PAS used for quantification in BED format
--model regression model
--factor_path 'regression/normalize_factor'
--depth default=1.
--out output of the regression results
--threshold Minimum RPM threshold for scaning item
Parallelization was implanted in APAIQ using multiple-processing in Python, for which DNA sequence and RNA-seq coverage across the whole genome were divided into hundreds of blocks. Thus, the run time and memory usage could be variable using different number of cores/CPU and different size of block (default is 100k bp). Usually, 2.5-4 G memory were required for each parallelized core/CPU.
Examples of input and output can be found under the 'demo' directory. A example of input was shown below.
chrX 20505 20601 0.00673982
chrX 20601 20605 0.0134796
chrX 20605 20701 0.00673982
chrX 20908 20980 0.00673982
chrX 20980 21008 0.0134796
chrX 21008 21080 0.00673982
chrX 21948 22047 0.00673982
chrX 22047 22048 0.0134796The first column is the chromosome. The 2nd and 3rd column are start and end of a genomic region. The fourth column indicate RNAseq coverage in the corresponding region. A exmaple of output was shown below.
#chromosme start end score id strand anno_id anno_source distance
chrX 303355 303356 34.55842697620392 chrX:+:16 + chrX:303356:+:PLCXD1 Gencode 0
chrX 1602523 1602524 44.11139154434204 chrX:+:26 + chrX:1602520:+:AKAP17A Gencode 4
chrX 2741311 2741312 35.39809334278107 chrX:+:33 + chrX:2741309:+:CD99 Gencode 3
chrX 2882817 2882818 43.826815128326416 chrX:+:40 + chrX:2882818:+:GYG2 3'UTR(M) 0
chrX 9717311 9717312 15.25328278541565 chrX:+:47 + chrX:9717314:+:TBL1X 3'UTR(M) -2
chrX 9719649 9719650 16.97154474258423 chrX:+:53 + chrX:9719652:+:TBL1X 3'UTR(M) -2
chrX 9719739 9719740 35.48719763755798 chrX:+:54 + chrX:9719740:+:TBL1X Gencode 0
chrX 9948362 9948363 16.15381270647049 chrX:+:55 + chrX:9948359:+:SHROOM2 3'UTR(M) 4The first three columns indicate the genomic coordinate of the identified PAS. The 4th column indicates the score of the PAS and usually 12 was used as the threshhold to filter reliable PAS, the 5th column is the PAS id and the 6th column is the strand. The 7th column is the cloest PAS from the DB file and the 8th column is the information of PAS from the DB file. The 9th is the distance.