According to the central dogma of molecular biology, the genomic sequence of an eukaryotic gene is transformed into the corresponding protein by the transcription, post-transcriptional and translation processes. One of the important steps of the post-transcriptional process is the polyadenylation process. This process consists of two steps: cleavage near the 3' end of a pre-mRNA and addition of a polyA tail at the cleavage site or polyA site. DeepPASTA is a tool for predicting such polyA sites from sequence and RNA secondary structure data. The tool can also predict tissue-specific polyA sites from sequence and RNA secondary structure data. When two polyA sites of a gene (of a particular tissue) are given, the tool can predict relatively dominant polyA site. Moreover, when a polyA site of a gene (of a particular tissue) is given, it can predict whether that site is one of the most (absolutely) dominant polyA sites of the gene or not.
@article{10.1093/bioinformatics/btz283,
author = {Arefeen, Ashraful and Xiao, Xinshu and Jiang, Tao},
title = "{DeepPASTA: deep neural network based polyadenylation site analysis}",
journal = {Bioinformatics},
volume = {35},
number = {22},
pages = {4577-4585},
year = {2019},
month = {04},
issn = {1367-4803},
doi = {10.1093/bioinformatics/btz283},
url = {https://doi.org/10.1093/bioinformatics/btz283},
eprint = {https://academic.oup.com/bioinformatics/article-pdf/35/22/4577/30706799/btz283.pdf},
}
-
The tool runs on linux machine.
-
After installing Anaconda2, please use DeepPASTA_env1.yml file to create DeepPASTA suitable environment. Please run the following command to activate the suitable environment:
conda env create -f DeepPASTA_env1.yml source activate DeepPASTA_envNote: If the anaconda environment creates problem, please follow the steps written in "Creating environment for DeepPASTA" section.
-
Please download the trained parameters and put them in respective folders.
In order to predict polyA sites, please use DeepPASTA_polyA_site_prediction_testing.py of polyA_site_prediction folder. Sample input files are given in the sample_input directory. Commands to run the polyA site prediction tool:
USAGE
cd polyA_site_prediction
python DeepPASTA_polyA_site_prediction_testing.py {OPTIONS}
OPTIONS
-testSeq <input_sequence_file> A FASTA file that contains human genomic sequences of length 200 nts.
-testSS <input_RNA_secondary_structure_file> An input file that contains the RNA secondary structures of the input sequences.
The tool expects three most energy efficient RNA secondary structures for each input sequence.
These RNA secondary structures are generated using [RNAshapes](https://academic.oup.com/bioinformatics/article/22/4/500/184565).
-o <output_file_name> Output file name is given using this option. If this option is not used then the tool outputs
AUC and AUPRC values of the prediction. In order to calculate the AUC and AUPRC values the tool
needs ground truth data. The ground truth data is added at the end of the title of each
sequence. E.g. for a positive sequence example, the title is >chr15_100354095_positive_1; on
the other hand, the title of a negative sequence example is >chr15_100565120_positive_0.
EXAMPLE
python DeepPASTA_polyA_site_prediction_testing.py -testSeq sample_sequence_input.hg19.fa -testSS sample_secondary_structure_input.txt
The model takes two files as input: sequence and RNA secondary structure files. The sequence file is a FASTA file that contains two lines per example. The first line is the title of the sequence and the second line is the 200 nts sequence. The RNA secondary structure has four lines per example. The first line is the title and the next three lines for three RNA secondary structures. The model outputs AUC and AUPRC values when -o option is not used. In order to get the AUC and AUPRC values, the user must give the ground truth values using the title. E.g. title_ground_truth_value; for a positive sequence example, the title is >chr15_100354095_positive_1; on the other hand, the title of a negative sequence example is >chr15_100565120_positive_0. If the user uses -o option, the model will output the predicted likelihood values in an output file.
In order to predict tissue-specific polyA sites, please use DeepPASTA_tissue_specific_polyA_site_prediction_testing.py of tissue_specific_polyA_site_prediction folder. Sample input files are given in the sample_input directory. Commands to run the tissue-specific polyA site prediction tool:
USAGE
cd tissue_specific_polyA_site_prediction
python DeepPASTA_tissue_specific_polyA_site_prediction_testing.py {OPTIONS}
OPTIONS
-test <input_sequence_file> A FASTA file that contains human genomic sequences of length 200 nts.
-tests <input_RNA_secondary_structure_file> An input file that contains the RNA secondary structures of the input sequences.
The tool expects three most energy efficient RNA secondary structures for each input sequence.
These RNA secondary structures are generated using [RNAshapes](https://academic.oup.com/bioinformatics/article/22/4/500/184565).
-testl <ground_truth_label_file> An input file for the ground truth labels. The ground truth labels help the tool to calculate AUC and AUPRC values.
-o <output_file_name> Output file name is given using this option. This option prints the result in an output file.
EXAMPLE
python DeepPASTA_tissue_specific_polyA_site_prediction_testing.py -test sample_sequence_input.hg19.fa -tests sample_secondary_structure_input.txt -testl sample_tissue_specific_label.txt
The model takes two files: sequence and RNA secondary structure files. The sequence file is a FASTA file that contains two lines per example. The first line is the title of the sequence and the second line is the 200 nts sequence. The RNA secondary structure has four lines per example. The first line is the title and the next three lines for three RNA secondary structures. The model outputs AUC and AUPRC values when -testl option is used. Using -testl option the user have to give the ground truth data. For each example, the ground truth data has two lines: these two lines are title and read counts (of nine tissues separated by comma), respectively. If -o option is used the model outputs tissue-specific polyA site prediction in a file. For each input example, the output file has a line containing title and nine likelihood values (separated by comma) for nine tissues.
The tool can also predict tissue-specific relatively dominant polyA sites. The files to run the tissue-specific relatively dominant polyA site prediction model are in tissue_specific_relatively_dominant folder. For an example, if an user wants to run the liver tissue relatively dominant polyA site prediction model, he/she have to follow the following commands:
USAGE
cd tissue_specific_relatively_dominant/tissue_set_one/liver
python DeepPASTA_relatively_dominant_liver_testing.py {OPTIONS}
OPTIONS
-test <input_file> An input file that contains the gene name, two polyA sites, two sequences around the polyA sites, the sequences
corresponding RNA secondary structures. If the user wants to caculate the AUC and AUPRC values of the prediction,
he/she also have to provide the polyA sites corresponding read counts in this file.
-o <output_file_name> This option prints the result in an output file.
EXAMPLE
python DeepPASTA_relatively_dominant_liver_testing.py -test sample_relatively_dominant_input_liver.txt
If the user wants to calculate the AUC and AUPRC values of the prediction, he/she must input a file that contains the gene name, first polyA site location, second polyA site location, sequence (200 nts) around the first polyA site, sequence (200 nts) around the second polyA site, read count of the first polyA site, read count of the second polyA site, RNA secondary structure of the first sequence, and RNA secondary structure of the second sequence. If the user wants to output the prediction result to a file using -o option, the input file must contains all the above information except the read counts. The output file contains the gene name, the two polyA site locations and the probabilities of relative dominance of these two polyA sites. Sample input files of the tissue-specific relatively dominant polyA site prediction model are given in the sample_input folder.
The tool can predict tissue-specific absolutely dominant polyA site when sequence (and RNA secondary structure) around a polyA site of a gene is given as input. The files to run the tissue-specific absolutely dominant polyA site prediction model are in tissue_specific_absolutely_dominant folder. For an example, if an user wants to run the liver tissue absolutely dominant polyA site prediction model, he/she have to follow the following commands:
USAGE
cd tissue_specific_absolutely_dominant/liver
python DeepPASTA_absolutely_dominant_liver_testing.py {OPTIONS}
OPTOINS
-test <input_file> An input file that contains the gene name, tissue name, polyA site location, sequence around the polyA site, RNA secondary structure
of the sequence. If the user wants to calculte the AUC and AUPRC values of the prediction, he/she have to provide the
label of the polyA site in this file.
-o <output_file_name> This option prints the result in an output file.
EXAMPLE
python DeepPASTA_absolutely_dominant_liver_testing.py -test sample_absolutely_dominant_input_liver.txt
If the user wants to calculate the AUC and AUPRC values of the prediction, he/she must input a file that contains the gene name, tissue name, polyA site location, sequence (200 nts) around the polyA site, RNA secondary structure of the sequence, and absolutely dominant label. If the user wants to output the prediction result to a file using -o option, the input file must contains all the above information except the label. The output file contains the gene name, polyA site location and likelihood value of the prediction. Sample input files of the tissue-specific absolutely dominant polyA site prediction model are given in the sample_input folder.
After installing anaconda2, please run the following commands to create the environment for DeepPASTA:
conda create -n DeepPASTA_env python=2.7.5
source activate DeepPASTA_env
conda install -c anaconda keras
conda install -c anaconda scikit-learn
The first step of generating secondary structures uses RNAshapes (Steffen et al., 2005). The output from RNAshapes is then converted to DeepPASTA suitable format using a sequence of commands. The necessary commands to generate DeepPASTA suitable secondary structures from sequences are given below:
./RNAshapes -f <input_file_in_fa_format> -s -c 5 -t 1 -w 100 -W 100 -O 'D{%s\n}' > <output_file_in_txt_format>
python combining_substructure.py -i <output_from_the_previous_command> -o <output_file_in_txt_format>
python filtering_number_of_ss.py -n 3 -i <output_from_the_previous_command> -o <output_file_in_txt_format>
python shape_assign_per_nucleotide.py -c 3 -i <output_from_the_previous_command> -o <output_file_in_txt_format>
RNAshapes (Steffen et al., 2005) is provided in the generating_secondary_structure_from_sequence directory for user convenience. Please follow the manual of RNAshapes (Steffen et al., 2005) for the possible options. In filtering_number_of_ss.py and shape_assign_per_nucleotide.py scripts, 3 is used to generate the three most probable secondary structreus for a given sequence.
We have used DeepPASTA to perform a genome-wide polyA site prediction for human based on the PolyA-Seq data in (Derti et al., 2012). The prediction results can be found here.
- Derti, A. et al., (2012) A quantitative atlas of polyadenylation in five mammals. Genome Research, 22 (6), 1173-1183
- Steffen, P. et al., (2005) RNAshapes: an integrated RNA analysis package based on abstract shapes. Bioinformatics, 22 (4), 500-503
Note: If you have any question or suggestion please feel free to email: aaref001@ucr.edu or ashraful.arefeen@csebuet.org