UC San Diego

Accurate genome-scale identification of somatic mutations in single mammalian cells remains very challenging. Current single-cell genome sequencing approaches generate tens of thousands of false positive calls per genome. This is manageable for calling the millions of germline variants or single nucleotide polymorphisms (SNP).However, the number of false positives could greatly outnumber that of true somatic mutations per genome, resulting in an unacceptable level of false discovery rate and limiting the clinical application of single-cell genome sequencing. In this dissertation, I describe a strategy called SISSOR (SIngle-Stranded Sequencing in micrOfluidic Reactors) to overcome this limitation. A microfluidic processor was designed to enable the separation of long single-stranded DNA strands in a single mammalian cell, random partitioning of megabase-size single-stranded molecules into multiple nanoliter-size micro-reactors for unbiased amplification, and sequencing library construction. By separating, amplifying and sequencing megabase-size Watson and Crick DNA strands of the homologous chromosome pairs in a single cell, potential errors due to amplification and sequencing can be removed using the consensus calls on the two complementary strands, and long-range haplotype assembly can also be obtained. Using the microfluidic processor, I implemented the strategy by actual amplification and sequencing of the genomes for three single cells. I demonstrated that sequencing accuracy can be improved by two orders of magnitude and the length of haplotype assembly can also be dramatically increased as compared to other currently available methods.