The Knowledge Exchange video series had one on MinKNOW features and Updates that I watched tonight. MinKNOW has evolved in the past couple of years! Rich Roman, a Technical Product Manager from ONT, described the changes in Model Selection and incorporation of the Flip Flop Basecalling model. The high accuracy (HAC) basecalling model contains a computationally intense flip flop architecture and higher accuracy basecalling. The fast basecalling model is a “light version” of the flip flop algorithm. The modified basecalling model detects methylation based on different E. coli and human data/models. Previously, one fast5 file per read was created. The default is now set to 4,000 reads per file to balance number of files and file size. Now, Pod5 files are used. Fast5 and Fastq files are compressed with VBZ and gzip compression. They noted that 100 Gbases of data result in 2 Tb of uncompressed Fast5! With compression, that space usage is reduced to 740 Gb. Barcode demultiplexing was added in the 19.11 release and included a graph of barcode accumulation. Rebasecalling can be performed with the Analysis tab on MinKNOW. Run reports summarize software versions and all output graphs.
