CNV-P is a novel and post–processing approach for CNV filtering.
Cite this as
Wang T, Sun J, Zhang X, Wang W, Zhou Q. 2021. CNV-P: a machine-learning framework for predicting high confident copy number variations. PeerJ 9:e12564 https://doi.org/10.7717/peerj.12564
python3
sklearn
matplotlib
pysam
pandas
numpy
conda install python=3.7.0
conda install -c anaconda scikit-learn=0.21
conda install -c conda-forge matplotlib
conda install -c bioconda pysam
conda install -c anaconda pandas
conda install -c anaconda numpy
run the "python script/CNV-P_predict_main.py -h" to see the USAGE;
usage: CNV-P_predict_main.py [-h] [-m model] -b bamfile -s CNVcaller -bed
BEDfile -bas basfile [-sam Samplename]
[-o outdir]
optional arguments:
-h, --help show this help message and exit
-m model, --model model
which model you want to used
-b bamfile, --bam bamfile
file provides features
-s CNVcaller, --soft CNVcaller
which CNV caller you used
-bed bedfile, --CNV_bed bedfile
the format of input file
-bas basfile, --basfile basfile
file that provide mean insert-size and sequencing
depth
-n Samplename, --Samplename Samplename
Samplename that use as prefix of result
-o outdir, --output outdir
output directory
model: should be one of RF (Random Forest), GBC (Gradient Boosting classifier) and SVM (Support Vector Machine)
CNVcaller: Lumpy, Manta, Pindel, Delly and breakdancer is currently supported, for Other software needs to be pre-trained(see 2.training for other CNV callers)
bamfile: BAM file should generated by a read aligner that supports partial read alignments, such as BWA-MEM
bedfile: This file should be 5 Columns: chromsome, start, end, length of CNV, type of CNV (DUP:1,DEL:0)
for example (test_data/HG002.Lumpy.fil.mer.bed):
chr19 350768 351961 1194 1
chr19 434243 434587 345 1
chr19 566222 569347 3126 0
chr19 878739 879857 1119 1
chr19 1182660 1183097 438 0
chr19 1572816 1573149 334 0
chr19 2033040 2033182 143 0
chr19 2713161 2714159 999 0
basfile: this file should be 4 columns: Samplename, median value of insert size, standard deviation of insert size, coverage
for example (test_data/HG002.bam.bas):
Samplename median_insert_size insert_size_median_sd coverage
HG002 568.177944 163.819637 35.41
samplename.feature.txt: Extracted feature matrix.
samplename.pre.prop.txt: The prediction result and probability score. Including 7 columns:
ChrID: Chromosome (e.g. chr3, chrY)
start: Start coordinate on the chromosome
end: End coordinate on the chromosome
length: length of CNV
CNV_type: type of CNV (DUP:1,DEL:0)
class: predicting results (true CNV：1 ,false CNV: 0)
probability_score: Probability of this CNV to be true
python script/CNV-P_predict_main.py -m RF -b Test_data/HG002.test.bam -s Lumpy -bed Test_data/HG002.Lumpy.fil.mer.bed -bas Test_data/HG002.bam.bas -sam HG002 -o Test_data/out/
For training a model for other CNV callers, use 'CNV-P_featureExtract_main.py' to perform features extraction:
python script/CNV-P_featureExtract_main.py -b test-data/HG002.test.bam -bed test-data/HG002.Lumpy.fil.mer.bed -bas test-data/HG002.bam.bas -sam HG002 -o test-data/out/
then，run the "script/CNV-P_training_main.py" to train a model
run " python script/CNV-P_training_main.py -h " to see the USAGE;
usage: CNV-P_training_main.py [-h] [-m model] -s CNVcaller -fea featuresfile
-lab labelfile [-o outdir]
optional arguments:
-h, --help show this help message and exit
-m model, --model model
which model you want to used
-s CNVcaller, --soft CNVcaller
which CNV caller you used
-fea featuresfile, --features featuresfile
file that provide traing features
-lab labelfile, --labelfile labelfile
file that provide CNV label, true CNVs labeled as
1,false CNVs labeled as 0, The order should
corresponds to CNV_bed file(-bed/--CNV_bed) one to one
-o outdir, --output outdir
output directory
featuresfile: file that provide traing features, results from 'CNV-P_featureExtract_main.py'
labelfile: one column, true CNVs labeled as 1,false CNVs labeled as 0
for example (see test-data/HG002.Lumpy.chr1.label.txt):
1
1
0
0
0
0
1
0
1
CNVcaller.model.train_model.m: the classifier you trained
CNV-P_CNVcaller_model_Classifier.ROC.pdf, CNV-P_CNVcaller_model_Classifier.ROC.png: the ROC of 10fold-cross_validation
python script/CNV-P_training_main.py -s Lumpy -fea test-data/HG002.Lumpy.chr1.feature.txt -lab test-data/HG002.Lumpy.chr1.label.txt -o test-data/out/
Please help us improve CNV-P by reporting bugs or ideas on how to make things better.
We compared the performance of CNV-P with that of CNV- JACG (Zhuang et al. 2020), MetaSV (Mohiyuddin et al. 2015) and hard cutoff method in the same datasets. Since MetaSV currently does not support Delly's output, only four CNV detection tools (Lumpy, Manta, Pindel, and breakdancer) were taken into consideration. CNV-JACG was conducted running with default parameters. MetaSV was carried out with complete mode. For hard cutoff method, we used SR and RP as the evidence to support the existence of CNVs, therefore, the number of SR and RP greater than 2, 5, and 10 were set as hard cutoff to evaluate. SURVIVOR(Jeffares et al. 2017) was used to merge fragments with 80% overlap after filtering by CNV-P, CNV- JACG, MetaSV and hard cutoff method.
Process framework：
Comparison with CNV-JACG, MetaSV and hard cutoff method in NA12878 and HG002.
| Sample | method | precision | recall | F1-score |
|---|---|---|---|---|
| NA12878 | RAW | 0.6032 | 1.0000 | 0.7525 |
| Hard_Cutoff_2 | 0.6197 | 0.9792 | 0.7590 | |
| Hard_Cutoff_5 | 0.7145 | 0.8630 | 0.7818 | |
| Hard_Cutoff_10 | 0.7780 | 0.6976 | 0.7356 | |
| CNV-JACG | 0.6828 | 0.7496 | 0.7146 | |
| MetaSV | 0.7094 | 0.8817 | 0.7862 | |
| CNV-P | 0.9007 | 0.7977 | 0.8461 | |
| HG002 | RAW | 0.2054 | 1.0000 | 0.3408 |
| Hard_Cutoff_2 | 0.4026 | 0.9729 | 0.5695 | |
| Hard_Cutoff_5 | 0.5740 | 0.8653 | 0.6901 | |
| Hard_Cutoff_10 | 0.6642 | 0.7482 | 0.7037 | |
| CNV-JACG | 0.5443 | 0.7076 | 0.6153 | |
| MetaSV | 0.5917 | 0.8274 | 0.6900 | |
| CNV-P | 0.7078 | 0.7516 | 0.7290 |