A framework and an algorithm to detect low-abundance DNA by a handy sequencer and a palm-sized computer

Motivation: Detection of DNA at low abundance with respect to the entire sample is an important problem in areas such as epidemiology and field research, as these samples are highly contaminated with non-target DNA. To solve this problem, many methods have been developed to date, but all require additional time-consuming and costly procedures. Meanwhile, the MinION sequencer developed by Oxford Nanopore Technology (ONT) is considered a powerful tool for tackling this problem, as it allows selective sequencing of target DNA. The main technology employed involves rejection of an undesirable read from a specific pore by inverting the voltage of that pore, which is referred to as 'Read Until'. Despite its usefulness, several issues remain to be solved in real situations. First, limited computational resources are available in field research and epidemiological applications. In addition, a high-speed online classification algorithm is required to make a prompt decision. Lastly, the lack of a theoretical approach for modeling of selective sequencing makes it difficult to analyze and justify a given algorithm. Results: In this paper, we introduced a statistical model of selective sequencing, proposed an efficient constant-time classifier for any background DNA profile, and validated its optimal precision. To confirm the feasibility of the proposed method in practice, for a pre-recorded mock sample, we demonstrate that the method can selectively sequence a 100 kb region, consisting of 0.1% of the entire read pool, and achieve approximately 500-fold amplification. Furthermore, the algorithm is shown to process 26 queries per second with a $500 palm-sized next unit of computing box using an Intel (R) Core (TM) i7 CPU without extended computer resources such as a GPU or high-performance computing. Next, we prepared a mixed DNA pool composed of Saccharomyces cerevisiae and lambda phage, in which any 200 kb region of S.cerevisiae consists of 0.1% of the whole sample. From this sample, a 30-230 kb region of S.cerevisiae chromosome 1 was amplified approximately 30-fold. In addition, this method allowed on-the-fly changing of the amplified region according to the uncovered characteristics of a given DNA sample.