circFL-refine: a tool for full-length circRNA isoform construction based on convolutional neural network

Introduction

circFL-refine is a circRNA analysis pipeline to evaluate and correct full-length structure of circRNA isoform

Availability

circFL-refine is a free software, which can be downloaded from https://github.com/yangence/circFL-refine

Well trained human and mouse donor/acceptor models are available at https://github.com/yangence/circFL-refine/tree/master/model

Curated datasets of full-length circRNA isoforms are visalized at UCSC, including isoforms from human and mouse

Curated datasets of full-length circRNA isoforms are available at https://github.com/yangence/circFL-refine/tree/master/example

Full-length circRNA nanopore simulation data can be generated by circFL-simu

Softwares and packages dependencies

Python 3.x.x and corresponding versions of numpy, pandas, torch, pyfasta, scikit-learn, docopt.

Installation

Install latest release from pip

pip install circFL-refine

Install latest release from source codes

git clone https://github.com/yangence/circFL-refine.git
cd circFL-refine/script
pip install -r requirements.txt
python setup.py install

Required files

Users can prepare the external files under the following instructions:

Indexed genome fasta file

samtools faidx $genome

Examples

https://github.com/yangence/circFL-refine/tree/master/example

Nanopore reads simulation

https://github.com/yangence/circFL-refine/tree/master/simu

Usage

Usage: circFL-refine <command> [options]
Command:
    train             Train CNN models to predict donor and acceptor sites
    evaluate          Evaluate full-length circRNA isoforms
    correct           Correct mistaken circRNA isoforms

train

Usage: circFL-refine train -f circ -g genome [-l lines] [-e epoch] [-o output]

Options:
    -h --help                   Show help message.
    -v --version                Show version.
    -f circ                     BED format file contain full-length circRNA information.
    -g genome                   Fasta file of genome.
    -l lines=K                  Only use first K lines for training, e.g. 30000.
    -e epoch                    Number of epoch for training [default: 1].
    -o output                   Output dir [default: circFL_refine].

evaluate

Usage: circFL-refine evaluate -f circ -g genome [-d donor_model] [-a acceptor_model] [-o output]

Options:
    -h --help                   Show help message.
    -v --version                Show version.
    -f circ                     BED format file contain full-length circRNA information.
    -g genome                   Fasta file of genome.
    -d donor_model              Donor model file, output of train commond.
    -a acceptor_model           Acceptor model file, output of train commond.
    -o output                   Output dir [default: circFL_refine].

correct

Usage: circFL-refine correct -f circ -g genome -d donor_model -a acceptor_model [-e mistake] [-i dist] [-o output]

Options:
    -h --help                   Show help message.
    -v --version                Show version.
    -f circ                     Evaluation result of full-length circRNA, output of evaluate commond.
    -g genome                   Fasta file of genome.
    -d donor_model              Donor model file, output of train commond.
    -a acceptor_model           Acceptor model file, output of train commond.
    -e mistake=k                Isoforms with less than k mistaken splice site to be corrected [default: 1].
    -i dist                     Maximum distance of nearby predicted splice site [default: 20].
    -o output                   Output dir [default: circFL_refine].

Output files

File name	Details
acceptor_model.net	acceptor splice site prediction model from train
donor_model.net	donor splice site prediction model from train
fl_evaluate.txt	evaluate results
fl_correct.detail	details of correct results
fl_correct.bed	corrected BED file from correct

fl_evaluate.txt

No.	Column name	Details
1	chrom	chromosome
2	chromStart	start coordinate of circRNA
3	chromEnd	end coordinate of circRNA
4	name	isoform ID
5	score	0
6	strand	strand of circRNA
7	thickStart	.
8	thickEnd	.
9	itemRgb	.
10	blockCount	number of exons in circRNA
11	blockSizes	block size of each exon
12	blockStarts	block start of each exon
13	donorSite	corrdinate of donor splice site
14	donorMotif	motif of donor splice site
15	donorPredict	prediction score of donor splice site
16	donorPredictMin	minium prediction score of donor splice site
17	acceptorSite	corrdinate of acceptor splice site
18	acceptorMotif	motif of acceptor splice site
19	acceptorPredict	prediction score of acceptor splice site
20	acceptorPredictMin	minium prediction score of acceptor splice site

Isoforms with donorPredictMin>0.5 and acceptorPredictMin>0.5 are considered as evaluate passed.

fl_correct.detail

No.	Column name	Details
1	chrom	chromosome
2	chromStart	start coordinate of circRNA
3	chromEnd	end coordinate of circRNA
4	name	isoform ID
5	score	0
6	strand	strand of circRNA
7	thickStart	.
8	thickEnd	.
9	itemRgb	.
10	blockCount	number of exons in circRNA
11	blockSizes	block size of each exon
12	blockStarts	block start of each exon
13	donorSite	corrdinate of donor splice site
14	donorMotif	motif of donor splice site
15	donorPredict	prediction score of donor splice site
16	donorPredictMin	minium prediction score of donor splice site
17	acceptorSite	corrdinate of acceptor splice site
18	acceptorMotif	motif of acceptor splice site
19	acceptorPredict	prediction score of acceptor splice site
20	acceptorPredictMin	minium prediction score of acceptor splice site
21	correct_donorSite	corrdinate of corrected donor splice site
22	correct_acceptorSite	corrdinate of corrected acceptor splice site
23	dN	number of mistaken donor splice site
24	aN	number of mistaken acceptor splice site
25	dM	maximum distance of corrected donor splice site to original site
26	aM	maximum distance of corrected acceptor splice site to original site

fl_correct.bed

No.	Column name	Details
1	chrom	chromosome
2	chromStart	start coordinate of circRNA
3	chromEnd	end coordinate of circRNA
4	name	isoform ID
5	score	0
6	strand	strand of circRNA
7	thickStart	.
8	thickEnd	.
9	itemRgb	.
10	blockCount	number of exons in circRNA
11	blockSizes	block size of each exon
12	blockStarts	block start of each exon

Copyright (C) 2021 Zelin Liu ([email protected]). See the LICENSE file for license rights and limitations.