Abstract:Background: As high throughput sequencing (HTS) technique advances, it has been widely used in cancer research. More specifically, HTS enables genome-wide detection of somatic alterations which implicate the dynamic of cancer development as well as therapeutic strategy. However, a number of complications, such as sample tissue lesion, imperfect experimental procedure and nucleotide polymorphism between individual, often induce non-somatic noise and jeopardize the detection of somatic alterations. We present a validated computational pipeline that allows accurate somatic alteration identification. Method: We created a number of cell line mixture samples and sequenced at averagely 1500x depth. Each mixture sample was composed of cell lines with pre-specified proportion in order to represent distinct tumor purity and clonality. Prior to mixture creation, variant candidates in each cell line sample were labeled as one of three categories: germline variant of high confidence, artifact or blacklist (excluded from future assessment). The categorization was based on HaplotypeCaller results followed by a series of quality control. The allele frequency distribution of germline variants from each cell line was validated in mixture samples. We used MuTect as prototype pipeline and implemented further optimization on a number of parameter settings in a supervised fashion. Lastly, the optimized pipeline was evaluated using an independent dataset and its performance was compared with a number of alternative variant callers. Result: Our benchmark showed that default parameter setting of ‘alt_allele_in_normal' and ‘nearby_gap_events' in MuTect were over-strict. 73% of variants rejected by these parameters were actually real somatic variants. We believed that, instead of using constant as parameter settings, the optimum parameters should be trained on data with similar sequencing depth. We therefore re-configured the parameters using logistic regression. Furthermore, minimum supporting depth and minimum allele frequency were determined by AUROC. With optimized pipeline, we were able to achieve 96% sensitivity and 2/M false positive rate (FPR) compared with 90% sensitivity and 8/M FPR using default parameters. We further tested the optimized pipeline in an independent dataset and compared the performance with several well-known variant callers including MuSE, VarScan2, Strelka, LoFreq, VarDict and MuTect2. The performance of each caller was ranked as following: our optimized pipeline > MuTect > MuTect2/Strelka/LoFreq > Muse > VarScan2/VarDict. Citation Format: Cheng Yan, Yingjiang Ye, Weihua Guo, Junhui Yang, Yufei Yang. An improved computational pipeline for tumor somatic alterations detection [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr 151.

Somatic Mutation Detection Using Ensemble of Machine Learning

Detecting Somatic Mutations Without Matched Normal Samples Using Long Reads

Cancer classification and pathway discovery using non-negative matrix factorization

Mitosomatic: a Tool for Accurate Identification of Mitochondrial DNA Somatic Mutations Without Paired Controls

Tsnad: an Integrated Software for Cancer Somatic Mutation and Tumour-Specific Neoantigen Detection

Integrating Somatic Mutations for Breast Cancer Survival Prediction Using Machine Learning Methods

Detecting somatic point mutations in cancer genome sequencing data: a comparison of mutation callers

Comparative Study of Ensemble Learning Approaches in the Identification of Disease Mutations

Somatic Point Mutation Calling in Low Cellularity Tumors

DeepGene: an Advanced Cancer Type Classifier Based on Deep Learning and Somatic Point Mutations

Predictive Modeling of Novel Somatic Mutation Impacts on Cancer Prognosis: A Machine Learning Approach Using the COSMIC Database

Machine learning predictions improve identification of real-world cancer driver mutations

Machine learning models-based on integration of next-generation sequencing testing and tumor cell sizes improve subtype classification of mature B-cell neoplasms

Implementation of ensemble machine learning algorithms on exome datasets for predicting early diagnosis of cancers

Abstract 151: an Improved Computational Pipeline for Tumor Somatic Alterations Detection

Deep learning-based approach to reveal tumor mutational burden status from whole slide images across multiple cancer types

Single-molecule genome-wide mutation profiles of cell-free DNA for non-invasive detection of cancer

Predicting loss-of-function impact of genetic mutations: a machine learning approach

Predicting Cancer Tissue-of-Origin by a Machine Learning Method Using DNA Somatic Mutation Data.

MSINGB: A Novel Computational Method Based on NGBoost for Identifying Microsatellite Instability Status from Tumor Mutation Annotation Data

Using the "Hidden" Genome to Improve Classification of Cancer Types