SNPmanifold: detecting single-cell clonality and lineages from single-nucleotide variants using binomial variational autoencoder

Abstract

Single-nucleotide-variant (SNV) clone assignment of high-covariance single-cell lineage tracing data remains a challenge due to hierarchical mutation structure and many missing signals. We develop SNPmanifold, a Python package that learns an SNV embedding manifold using a binomial variational autoencoder to give an efficient and interpretable cell-cell distance metric. We demonstrate that SNPmanifold is a suitable tool for analysis of complex, single-cell SNV mutation data, such as in the context of demultiplexing a large number of donors and somatic lineage tracing via mitochondrial SNV data and can reveal insights into single-cell clonality and lineages more accurately and comprehensively than existing methods.