NCM 2022: The Dorado basecaller — a deep dive

A basecaller maps raw signal, measured from sensors in a flow cell, to sequences of nucleotides. This involves several steps, such as scaling, chunking, encoding, decoding, and stitching, each of which presents computational challenges at the scale of tens of millions of samples per second. This talk will describe these steps and explain some of the computational advances we have achieved in the Dorado basecaller to allow these models to run at high speed.