A universal SNP and small-indel variant caller using deep neural networks

Poplin, Ryan; Chang, Pi-Chuan; Alexander, David H.; Schwartz, Scott; Colthurst, Thomas; Ku, Alexander; Newburger, Dan; Dijamco, Jojo; Nguyen, Nam V.; Afshar, Pegah Tootoonchi; Gross, Sam; Dorfman, Lizzie; McLean, Cory Y.; DePristo, Mark A.

doi:10.1038/nbt.4235

Cited by 1,145 publications

(925 citation statements)

References 33 publications

Supporting

Mentioning

917

Contrasting

Unclassified

Order By: Relevance

“…If a user wants to filter out FPs exhaustively without consideration of the loss of TPs, he/she can use FPfilter together with GATK-HF. Moreover, compared with GATK-HF, FPfilter is much less aggressive on filtering FPs in SNPs but more aggressive on filtering FPs in INDELs (Supplemental Table 5 We also noticed that other variant calling methods, such as VariantMetaCaller [17], freebayes [18] and DeepVariant [19], requiring little post-call filtering, were stated to perform better than GATK. But before their replacement of GATK, imporving the post-call filtering of GATK is still meaningful.…”

Section: Conclusion and Discussionmentioning

confidence: 98%

FPfilter: A false-positive-specific filter for whole-genome sequencing variant calling from GATK

Tan

Zhang

Yang

et al. 2020

Preprint

View full text Add to dashboard Cite

Tel/ Fax: +86-020-85222787 1. Abstract:Motivation: As whole genome sequencing (WGS) is becoming cost-effective progressivelly, it has been applied increasingly in medical and scientific fields.Although the traditional variant-calling pipeline (BWA+GATK) has very high accuracy, false positives (FPs) are still an unavoidable problem that might lead to unfavorable outcomes, especially in clinical applications. As a result, filtering out FPs is recommended after variant calling. However, loss of true positives (TPs) is inevitable in FP-filtering methods, such as GATK hard filtering (GATK-HF).Therefore, to minimize the loss of TPs and maximize the filtration of FPs, a comprehensive understanding of the features of TPs and FPs, and building an improved model of classification are necessary. To obtain information about TPs and FPs, we used Platinum Genome (PT) as the mutation reference and its 300× deep sequenced dataset NA12878 as the simulation template. Then random sampling across depth gradients from NA12878 was performed to study the depth effect.Results: FPs among heterozygous mutations were found to have pattern distinct from that of FPs among homozygous mutations. FPfilter makes use of this model to filter out FPs specifically. We evaluated FPfilter on a training dataset with depth gradients from NA12878 and a test dataset from NA12877 and NA24385. Compared with GATK-HF, FPfilter showed a significantly higher FP/TP filtration ratio and F-measure score. Our results indicate that FPfilter provides an improved model for distinguishing FPs from TPs and filters FPs with high specificity.Availability: FPfilter is freely available for download on GitHub (https://github.com/yuxiangtan/FPfilter). Users can easily install it from anaconda.

show abstract

Section: Conclusion and Discussionmentioning

confidence: 98%

FPfilter: A false-positive-specific filter for whole-genome sequencing variant calling from GATK

Tan

Zhang

Yang

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

“…Each Illumina and PacBio read is encoded into a multi-dimensional numerical array representing a sequence. Encoding of bases quality scores etc into numerical values, follows DeepVariant (Poplin, et al, 2018). When analyzing a site, these encoded read sequences are fed into the MoE, along with information on candidate alleles and their supporting reads.…”

Section: Moe Architecturementioning

confidence: 99%

“…Deep Neural Networks (DNNs) have recently become the preferred method for pattern recognition tasks and have been applied successfully to the problem of small variant calling (Poplin, et al, 2018). This was enabled by the availability of high-quality, benchmarked truth-sets from efforts such as the Genome-In-A-Bottle (GIAB) and Platinum Genomes (Eberle, et al, 2016).…”

Section: Introductionmentioning

confidence: 99%

HELLO: A hybrid variant calling approach

Ramachandran

Lumetta

Klee

et al. 2020

Preprint

View full text Add to dashboard Cite

Next Generation Sequencing (NGS) technologies that cost-effectively characterize genomic regions and identify sequence variations using short reads are the current standard for genome sequencing. However, calling small indels in low-complexity regions of the genome using NGS is challenging. Recent advances in Third Generation Sequencing (TGS) provide long reads, which call largestructural variants accurately. However, these reads have context-dependent indel errors in lowcomplexity regions, resulting in lower accuracy of small indel calls compared to NGS reads. When both small and large-structural variants need to be called, both NGS and TGS reads may be available.Integration of the two data types with unique error profiles could improve robustness of small variant calling in challenging cases. However, there isn't currently such a method integrating both types of data.We present a novel method that integrates NGS and TGS reads to call small variants. We leverage the Mixture of Experts paradigm which uses an ensemble of Deep Neural Networks (DNN), each processing a different data type to make predictions. We present improvements in our DNN design compared to previous work such as sequence processing using one-dimensional convolutions instead of image processing using two-dimensional convolutions and an algorithm to efficiently process sites with many variant candidates, which help us reduce computations. Using our method to integrate Illumina and PacBio reads, we find a reduction in the number of erroneous small variant calls of up to ~30%, compared to the state-of-the-art using only Illumina data. We also find improvements in calling small indels in low-complexity regions.

show abstract

“…An orthogonal approach to variant identification utilizes neural networks. [6][7][8] A neural network is a machine learning framework that captures complex feature dependencies for prediction and inference. Previous variant callers used curated features or images of DNA pileup plots as inputs for neural network variant callers.…”

Section: Introductionmentioning

confidence: 99%

Whole GenomeDe NovoVariant Identification with FreeBayes and Neural Network Approaches

Richter

et al. 2020

Preprint

View full text Add to dashboard Cite

Motivation: De novo variant (DNV) calling typically relies on heuristic filters intrinsic to specific platforms and variant calling algorithms. FreeBayes and neural network approaches have overcome this limitation for variant calling, and we implemented a similar approach for DNV identification. Results: We developed a DNV calling framework that uses Genome Analysis Toolkit (GATK), FreeBayes and a neural network trained on Integrative Genomics Viewer pile-up plots (IGVbot). We identified DNVs in 2,390 WGS trios and benchmarked results against heuristics based on GATK parameters. Results were validated in silico and with Sanger sequencing, with the latter showing true positive rates of 98.4% and 97.3% for SNVs and indels, respectively. Taken together we describe a scalable framework for DNV identification based on both FreeBayes and neural network methods. Availability: Source code and documentation are available at https://github.com/ShenLab/igvclassifier and https://github.com/frichter/dnv_pipeline under the MIT license. Contact:

show abstract

A universal SNP and small-indel variant caller using deep neural networks

Cited by 1,145 publications

References 33 publications

FPfilter: A false-positive-specific filter for whole-genome sequencing variant calling from GATK

FPfilter: A false-positive-specific filter for whole-genome sequencing variant calling from GATK

HELLO: A hybrid variant calling approach

Whole GenomeDe NovoVariant Identification with FreeBayes and Neural Network Approaches

Contact Info

Product

Resources

About