Reconstruction of influenza a virus variants from PacBio reads

Pacific Biosciences (PacBio) sequencing is providing thousands of reads with the length up to 10,000 bases. In most cases this length is enough to cover entire region of interest however this technology has high (≈ 15%) error rate. We propose a method for viral haplotype reconstruction generalizes k-means clustering with Hamming distance and capable of handling up to 25% random errors. When applied to PacBio reads from an Influenza A Virus (IAV) sample with ten variants, our method was able to reconstruct the four most frequent.