Nanopore Sequencing using a Hidden Markov Model for Base-Calling

Abstract

Nanopore-based DNA sequencing has many features which recommend it over current state of the art sequencing by synthesis methods; increased read length, reduced sample requirements, and increased speed to name a few. However, the accuracy of base-calling using the electrolytic current has thus far been relatively limited, from both solid-state and biological nanopores. Though part of this is due to the low signal-to-noise ratio, another significant contribution is the base resolution of the nanopore - it does not interrogate a single base at a time, rather the current is influenced by multiple bases at the same time.We suggest using this multi-base interrogation as an advantage rather than a disadvantage - each base in the DNA strand is read multiple times in this circumstance, allowing for improved accuracy. We have implemented a method to decode the electrical signature of 3bp resolution nanopore electrical measurements into a DNA sequence using a hidden Markov model. We produced simulated ionic current for all 64 possible triplets using atomic-resolution Brownian dynamics (BD).Using simulated current signatures, we have been able to demonstrate 98.3% base-calling accuracy for λ DNA, a substantially increased value compared to using only a single-base method (47.1%). When applied to 50kb fragments of the human genome, we found a similar median accuracy level (98.2%). Furthermore, we determined that there is a correlation between the local complexity of the sequence, as measured by Shannon's entropy, and the error rate; lower complexity sequence is more error-prone. Longer sequenced fragments have lower error rate - the more information input into the Markov model, the more effective the algorithm is at decoding the DNA sequence; this is in marked contrast to the dephasing problems endemic to current sequencing by synthesis methods.