Machine learning guided signal enrichment for ultrasensitive plasma tumor burden monitoring

Abstract

In solid tumor oncology, circulating tumor DNA (ctDNA) is poised to transform care through accurate assessment of minimal residual disease (MRD) and therapeutic response monitoring. To overcome the sparsity of ctDNA fragments in low tumor fraction (TF) settings and increase MRD sensitivity, we previously leveraged genome-wide mutational integration through plasma whole genome sequencing (WGS). We now introduce MRD-EDGE, a composite machine learning-guided WGS ctDNA single nucleotide variant (SNV) and copy number variant (CNV) detection platform designed to increase signal enrichment. MRD-EDGE uses deep learning and a ctDNA-specific feature space to increase SNV signal to noise enrichment in WGS by 300X compared to our previous noise suppression platform MRDetect. MRD-EDGE also reduces the degree of aneuploidy needed for ultrasensitive CNV detection through WGS from 1Gb to 200Mb, thereby expanding its applicability to a wider range of solid tumors. We harness the improved performance to track changes in tumor burden in response to neoadjuvant immunotherapy in non-small cell lung cancer and demonstrate ctDNA shedding in precancerous colorectal adenomas. Finally, the radical signal to noise enrichment in MRD-EDGE enables de novo mutation calling in melanoma without matched tumor, yielding clinically informative TF monitoring for patients on immune checkpoint inhibition.

Detecting cell-of-origin and cancer-specific methylation features of cell-free DNA from Nanopore sequencing

“…The cost per base of Nanopore sequencing is currently several-fold higher than Illumina, although the new generation of ONT PromethION sequencers is meant to address this and increase throughput. Single-nucleotide and indel error rates are higher for the ONT platform, which could pose an issue for whole-genome analysis of mutational signatures [39,40], something we do not investigate here but is theoretically possible from cfNano. While Nanopore error rates have improved significantly over the past several years, this is a weak point that should be considered if mutations are a priority.…”

Whole genome error-corrected sequencing for sensitive circulating tumor DNA cancer monitoring

“…We admixed reads from MEL-01 (stage IV melanoma patient, TF = 22%) and CTRL-05 (no known cancer) at different ratios to create admixtures of tumor fractions ranging from 10 −7 -10 −2 (n = 50 technical replicates per admixture) at 80x sequencing depth. Our group previously developed MRD-EDGE SNV 40 , a cancer-specific deep learning classifier that uses mutation sequence context and other features to analytically distinguish ctDNA from sequencing error at low tumor fractions. Our Ultima-specific denoising framework, in combination with the MRD-EDGE SNV deep-learning architecture, allowed ctDNA detection in the part per million range (Figure 1F), demonstrating the power of deeper WGS to increase ctDNA detection sensitivity.…”

“…Tumor fractions were estimated by dividing the number of filtered reads containing the somatic mutation by the total number of filtered reads. For the lower limit of detection estimation, denoised-reads were processed through MRD-EDGE 40 as described below.…”

Detecting cell-of-origin and cancer-specific methylation features of cell-free DNA from Nanopore sequencing

“…The cost per base of Nanopore sequencing is currently several-fold higher than Illumina, although the new generation of ONT PromethION sequencers are meant to address this as well as increase throughput. Single-nucleotide and indel error rates are higher for the ONT platform, which could pose an issue for whole-genome analysis of mutational signatures ( 38, 39 ), something we do not investigate here but is theoretically possible from cfNano. While Nanopore error rates have improved significantly over the past several years, this is a weak point that should be considered if mutations are a priority.…”