London Calling 2023: Nanomix: methylation-based cell-type deconvolution for low-pass nanopore sequencing

Deconvoluting heterogeneous DNA mixtures is typically accomplished using non-negative least squares (NNLS) when working with methylation arrays and short-read bisulfite sequencing. However, it is unclear how to best deconvolute mixtures using single-molecule base modification measurements. To address this, we developed Nanomix — a suite of tools designed for methylation deconvolution of nanopore sequencing data. Nanomix can simulate in silico mixtures of purified reference cell methylomes. We simulated lung and blood cell mixtures to benchmark deconvolution algorithms and reference cell atlases. We introduce new likelihood-based algorithms that demonstrate higher resilience to sequencing error and perform well at low coverage. Unlike previous methods, our tool can assign individual reads to cell types. Our results show how nanopore sequencing can leverage native methylation calling to deconvolute a complex mixture of DNA, which has applications to tumour purity testing and cell-free DNA analysis of liquid biopsies.