DNA Sequencing using deep Recurrent Neural Networks

Sequencing by DNA translocation is an emerging technology which offers cheaper and faster devices for DNA sequencing. However, precisely determining the bases present in the sequence from noisy and lengthy electric signals is particu larly challenging and offers an interesting research problem. In this project two basecalling methods DeepNano and Chiron are evaluated and the fundamental advantages and drawbacks associated with them are analyzed. Chiron uses a combination of CNN-RNN model as opposed to RNN in case of Deepnano. Comparison between the methods is made in the context of performance accuracy,speed, complexity and generalizability. Chiron with it’s complex model provides better read accuracy as well as generalizes well with unseen data when compared to Deepnano.

his project was undertaken as part of Genomic Signal Processing and Data Science course during Aug-Dec 2018. Further details about the project and results are available here.