Cancer type classification in liquid biopsies based on sparse mutational profiles enabled through data augmentation and integration

Abstract

Identifying the cell of origin of cancer is important to guide treatment decisions. However, in patients with 'cancer of unknown primary' (CUP), standard diagnostic tools often fail to identify the primary tumor. As an alternative, machine learning approaches have been proposed to classify the cell of origin based on somatic mutation profiles in the genome of solid tissue biopsies. However, solid biopsies can cause complications and certain tumors are not accessible. A promising alternative would be liquid biopsies, which contain ctDNA originating from the tumor. Problematically, somatic mutation profiles of tumors obtained from liquid biopsies are inherently extremely sparse and current machine learning models fail to perform in this setting. Here we propose an improved machine learning method to deal with the sparse nature of liquid biopsy data. Firstly, we downsample the SNVs in the samples in order to mimic sparse data conditions. Then extensive data augmentation is performed to artificially increase the number of training samples in order to enhance model robustness under sparse data conditions. Finally, we employ data integration to merge information from i) somatic single nucleotide variant (SNV) density across the genome, ii) somatic SNVs in driver genes and iii) trinucleotide motifs. Our adapted method achieves an average accuracy of 0.88 on the data where only 70% of SNVs are retained, which is comparable to an average accuracy of 0.87 with the original model on the full SNV data. Even when only 2% of the data is retained, the average accuracy is 0.65 compared to 0.41 with the original model. The method and results presented here open the way for application of machine learning in the detection of the cell of origin of cancer from sparse liquid biopsy data.

Mutation-Attention (MuAt): deep representation learning of somatic mutations for tumour typing and subtyping

“…Hence, MuAt results may be informative of tumour origins even if the prediction is not correct. We also observed relatively good performance with downsampled WGS data, suggesting MuAt may find use in low-coverage WGS data, pediatric tumours, and in cell-free DNA applications where only a fraction of the somatic mutation catalogue of a tumour might be captured [22].…”

“…Both Jiao et al and Salvadores et al [21] found the utility of driver mutations in accurately predicting tumour types to be limited due to the relatively small number of driver alterations per tumour, few recurrent driver alterations, and lack of strong tumour type specificity for cancer driver genes. Recently, Danyi et al showed data augmentation to be an effective strategy for tumour typing with sparse sequencing data such as sequencing of ctDNA [22].…”

“…Recently, Danyi et al . showed data augmentation to be an effective strategy for tumour typing with sparse sequencing data such as sequencing of ctDNA [22].…”

ctDNA to Guide Adjuvant Therapy in Localized Colorectal Cancer (CRC)

“…Machine learning approaches have been proposed to classify the cell of origin based on somatic mutation profiles in the genome of solid tissue biopsies. Therefore, it is crucial to investigate the applicability of sparse somatic mutation profiles in the identification of 'cell of origin' and explore potential improvements of the data analysis and prediction models to overcome sparsity [49].…”

Aggregation Tool for Genomic Concepts (ATGC): A deep learning framework for somatic mutations and other sparse genomic measures

“…A readily available label in cancer datasets is the cancer type, and this task can have practical importance for when the tumor of origin for a metastatic cancer cannot be determined 19 . The types of mutations of a cancer are influenced by the mutational processes of its etiology while the genomic distribution of its mutations is influenced by histology of origin, and these features of somatic mutations have been shown to be capable of classifying cancers [20][21][22][23] . However, these previous studies relied on manually generated features-the onehotting of mutational context and binning of genomic location.…”