Comprehensive identification of somatic nucleotide variants in human brain tissue

Abstract

Post-zygotic mutations incurred during DNA replication, DNA repair, and other cellular processes lead to somatic mosaicism. Somatic mosaicism is an established cause of various diseases, including cancers. However, detecting mosaic variants in DNA from non-cancerous somatic tissues poses significant challenges, particularly if the variants only are present in a small fraction of cells. Here, the Brain Somatic Mosaicism Network conducted a coordinated, multi-institutional study to: (i) examine the ability of existing methods to detect simulated somatic single nucleotide variants (SNVs) in DNA mixing experiments; (ii) generate multiple replicates of whole genome sequencing data from the dorsolateral prefrontal cortex, other brain regions, dura mater, and dural fibroblasts of a single neurotypical individual; (iii) devise strategies to discover somatic SNVs; and (iv) apply various approaches to validate somatic SNVs. These efforts led to the identification of 43 bona fide somatic SNVs that ranged in variant allele fractions from ~0.005 to ~0.28. Guided by these results, we devised best practices for calling mosaic SNVs from 250X whole genome sequencing data in the accessible portion of the human genome that achieve 90% specificity and sensitivity. Finally, we demonstrated that analysis of multiple bulk DNA samples from a single individual allows the reconstruction of early developmental cell lineage trees. Thus, this study provides a unified set of best practices to detect somatic SNVs in non-cancerous tissues. The data and methods are freely available to the scientific community and should serve as a guide to assess the contributions of somatic SNVs to neuropsychiatric diseases.

DeepMosaic: Control-independent mosaic single nucleotide variant detection using deep convolutional neural networks

“…400 variants underwent multiple levels of computational validations including haplotype phasing, CNV exclusion, population shared exclusion, as well as experimental validation such as whole-genome single cell sequencing, Chromium Linked-read sequencing (10X Genomics), PCR amplicon sequencing, and droplet digital PCR. After validation, 43 true positive MVs and 357 false positive variants were determined as gold-standard evaluation set for low-fraction single nucleotide MVs from the 250x WGS data 22 . We extracted deep whole-genome sequences for those variants, labeled them accordingly and used them as gold standard validation set for model selection (Supplementary Fig.…”

“…To estimate the performance of the pre-trained models and select the model with the best performance for DeepMosaic-CM, we introduced an independent gold-standard dataset. Variants were computationally detected from replicated sequencing experiments generated from 6 distinct sequencing centers and validated in 5 different centers, known as the common reference tissue project from the Brain Somatic Mosaicism Network 23 . 400 variants underwent multiple levels of computational validations including haplotype phasing, CNV exclusion, population shared exclusion, as well as experimental validation such as whole-genome single cell sequencing, Chromium Linked-read sequencing (10X Genomics), PCR amplicon sequencing, and droplet digital PCR.…”

“…To compare the different models after training, we employed an independent gold-standard validation dataset of ~400 MVs from one brain sample 22 (Methods). On these, EfficientNet-b4 showed the highest accuracy, Matthews’s correlation coefficient, and true positive rate when trained for 6 epochs (Supplementary Fig.…”

“…Ten different convolutional neural network models were trained on 180,000 experimentally validated positive and negative biological variants from 29 WGS data from 6 individuals sequenced at 100x 16, 17 (BioData1), as well as simulated data with different AFs (SimData1) resampled to a different depth. Models were evaluated based upon an independent gold-standard biological dataset from the 250x WGS data of the Reference Tissue Project of the Brain Somatic Mosaicism Network 23 (BioData2) as well as an independent 300x WGS dataset from the Brain Somatic Mosaicism Network Capstone project 18 (Biodata3). DeepMosaic was further benchmarked on 16 independent biological datasets from 200x WGS data 27 (BioData4) as well as 619,740 independently simulated variants (SimData2 and SimData3).…”

“…Accuracy, Matthews’s correlation coefficient (MCC), and Sensitivity of different network structures trained with the same data with different epochs. EfficientNet-b4 trained at 6 epochs demonstrated the highest Accuracy, MCC, and Sensitivity on the gold standard validation set 23 (BioData2); thus it was used as the default core model for DeepMosaic. We additionally provide an option for experienced users to train their own models with self-labeled training data.…”