Background Accurate detection of somatic mutations is challenging but critical in understanding cancer formation, progression, and treatment. We recently proposed NeuSomatic, the first deep convolutional neural network-based somatic mutation detection approach, and demonstrated performance advantages on in silico data. Results In this study, we use the first comprehensive and well-characterized somatic reference data sets from the SEQC2 consortium to investigate best practices for using a deep learning framework in cancer mutation detection. Using the high-confidence somatic mutations established for a cancer cell line by the consortium, we identify the best strategy for building robust models on multiple data sets derived from samples representing real scenarios, for example, a model trained on a combination of real and spike-in mutations had the highest average performance. Conclusions The strategy identified in our study achieved high robustness across multiple sequencing technologies for fresh and FFPE DNA input, varying tumor/normal purities, and different coverages, with significant superiority over conventional detection approaches in general, as well as in challenging situations such as low coverage, low variant allele frequency, DNA damage, and difficult genomic regions

Background The use of a personalized haplotype-specific genome assembly, rather than an unrelated, mosaic genome like GRCh38, as a reference for detecting the full spectrum of somatic events from cancers has long been advocated but has never been explored in tumor-normal paired samples. Here, we provide the first demonstrated use of de novo assembled personalized genome as a reference for cancer mutation detection and quantifying the effects of the reference genomes on the accuracy of somatic mutation detection. Results We generate de novo assemblies of the first tumor-normal paired genomes, both nuclear and mitochondrial, derived from the same individual with triple negative breast cancer. The personalized genome was chromosomal scale, haplotype phased, and annotated. We demonstrate that it provides individual specific haplotypes for complex regions and medically relevant genes. We illustrate that the personalized genome reference not only improves read alignments for both short-read and long-read sequencing data but also ameliorates the detection accuracy of somatic SNVs and SVs. We identify the equivalent somatic mutation calls between two genome references and uncover novel somatic mutations only when personalized genome assembly is used as a reference. Conclusions Our findings demonstrate that use of a personalized genome with individual-specific haplotypes is essential for accurate detection of the full spectrum of somatic mutations in the paired tumor-normal samples. The unique resource and methodology established in this study will be beneficial to the development of precision oncology medicine not only for breast cancer, but also for other cancers.

Background The cancer genome is commonly altered with thousands of structural rearrangements including insertions, deletions, translocation, inversions, duplications, and copy number variations. Thus, structural variant (SV) characterization plays a paramount role in cancer target identification, oncology diagnostics, and personalized medicine. As part of the SEQC2 Consortium effort, the present study established and evaluated a consensus SV call set using a breast cancer reference cell line and matched normal control derived from the same donor, which were used in our companion benchmarking studies as reference samples. Results We systematically investigated somatic SVs in the reference cancer cell line by comparing to a matched normal cell line using multiple NGS platforms including Illumina short-read, 10X Genomics linked reads, PacBio long reads, Oxford Nanopore long reads, and high-throughput chromosome conformation capture (Hi-C). We established a consensus SV call set of a total of 1788 SVs including 717 deletions, 230 duplications, 551 insertions, 133 inversions, 146 translocations, and 11 breakends for the reference cancer cell line. To independently evaluate and cross-validate the accuracy of our consensus SV call set, we used orthogonal methods including PCR-based validation, Affymetrix arrays, Bionano optical mapping, and identification of fusion genes detected from RNA-seq. We evaluated the strengths and weaknesses of each NGS technology for SV determination, and our findings provide an actionable guide to improve cancer genome SV detection sensitivity and accuracy. Conclusions A high-confidence consensus SV call set was established for the reference cancer cell line. A large subset of the variants identified was validated by multiple orthogonal methods.

Background: Ivosidenib (Servier Pharmaceuticals LLC) and enasidenib (Celgene) are inhibitors of mutant IDH1 and IDH2, respectively, approved for the treatment of relapsed/refractory (R/R) AML with an IDH1 or IDH2 mutation (ivosidenib is also approved for front-line use in patients ≥ 75 years or with comorbidities). While these drugs were approved based on durable complete remission (CR) + CR with partial hematologic recovery (CRh) rate, only 23-43% of patients responded. Thus, there is a need for a biomarker to better predict response. We previously presented a genomic model of response to enasidenib using linear discriminate analysis (Ghazanchyan et al., ASH 2018). In the two-part study presented here, we improved this modeling by adding clinical data, using machine learning-based approaches to model response to both enasidenib and ivosidenib, and generated an independent validation cohort using clinical samples.