Hidden Markov Model for Base Calling of MinION Nanopore Reads

MinION je sekvencer treće generacije. Zbog svoje male veličine i prihvatljive cijene popularnost mu raste. Jedna od mana MinION-a, s obzirom na druge sekvencere, viši je postotak greške kod dekodiranja. Kako bi se greška smanjila razvijaju se različiti alati za prevođenje signala u baze. Signali koje proizvede MinION se grupiraju u događaje te se ti događaji prevode u niz bazi. Prvi open-source alat za prevođenje događaja u baze je alat Nanocall. Na temelju Nanocall-a razvijen je alat opisan u ovom radu. Alat koristi Skriveni Markovljev model. Događaji su predstavljeni izlazima iz modela dok su baze predstavljene skrivenim događajima. Za treniranje vjerojatnosti prijelaza između skrivenih stanja koristi se Baum-Welch algoritam. Za treniranje vjerojatnosti između skrivenog stanja i izlaza korišten je algoritam maksimizacije očekivanja. Viterbi algoritam korišten je za dekodiranje. U tijeku je dodavanje poboljšanja kako bi se povisio postotak točnosti.MinION, by Oxford Nanopore Technologies, is a third generation sequencer. Due to its small size and low cost it has become very popular. One of its disadvantages in comparison to more expensive and bigger sequencers is a higher error rate in basecalling. In order to reduce the error new basecallers are being developed. Signals received from the MinION are segmented into a list of events and are then translated into a nucleotide sequence by a basecaller. One of the basecallers is Nanocall, first open source basecaller. Based on the implementation of Nanocall, basecaller described in this work has been made. Implemented basecaller relies on Hidden Markov Model to represent the events as the observations and the nucleotides as the hidden states. It uses Baum-Welch algorithm to update the transition probabilities and Expectation Maximization to update the emission probabilities. Having a trained model, Viterbi algorithm is used to decode the sequence of events. Improvements of the basecaller are still being added in order to lower the error rate