identifying cancer patients from GC-patterned fragment ends of cell-free DNA

Abstract

One of the most intriguing characteristics of cell-free DNA (cfDNA) from plasma is the sequence at the ends of the fragments. Previous studies have shown that these end-sequences are somewhat different in cancer patients than in healthy individuals. While investigating this characteristic, we noticed that the bases at the 5-prime ends of a double-stranded fragment were highly correlated with the GC content of that particular fragment. This led us to develop a method, called MendSeqS (Modified End-based sequencing System), that incorporates the correlation between end-motifs and GC content into the analysis of shallow (0.5x) whole genome sequencing (WGS). MendSeqS was able to classify patients with a sensitivity of 96% at 98% specificity in a cohort comprised of 107 individuals evaluated in our laboratory (43 with cancer and 64 without). In cohorts evaluated in three other laboratories, comprising a total o4 401 individuals (193 with cancer and 208 without), MendSeqS achieved a sensitivity of 87% at 98% specificity. MendSeqS could in principle be combined with other methods of cfDNA analysis to enhance cancer detection.

Machine learning to detect the SINEs of cancer

“…We wondered whether the reduced representation of AluS elements in the cfDNA from patients with cancer was the result of some unknown bias in amplification efficiency or sequencing generated from the RealSeqS approach, which used a single primer pair. To address this question, WGS data from our recently published study that included 65 cfDNA samples from individuals without cancer and 58 samples from patients with cancer were evaluated (55 colorectum, 2 lung, and 1 pancreas) (table S8) (33). Cancer samples had a global reduction of AluS representation compared with controls (P < 2.6 × 10 −5 , one-sided t test; Fig.…”