DeepBinner on SQK-RLB001 - Rapid Low Input by PCR Barcoding Kit

[jnesme]

Dear Ryan, thanks a lot for all your very useful tools for Nanopore data handling.

I'm currently trying to demultiplex a MinION run (FLO-MIN106) barcoded with SQK-RLB001 - Rapid Low Input by PCR Barcoding Kit. Barcodes 03; 04; 05; 06; 07; 08; 09 were used (tot. 7 barcodes)

I gave a shot at deepBinner using the Rapid model and while a lot of reads are not getting classified, there is still a fair amount that got assigned a barcode using default threshold. I was wondering if those results can be trusted in anyway ? My strategy would have been to then basecall each fast5 barcode directories with demultiplexing and then concatenate fasta for which both methods agrees, as you're suggesting for maximum stringency.

As I'm assembling fosmid inserts, I'm not too much scared by read yield but very much by barcode misclassifications.

Barcode     Count
      1        45
      2       134
      3      5639
      4      1819
      5       889
      6      1754
      7      1042
      8      1782
      9      1060
     10       574
     11       494
     12       170
   none     10198


Found 66,997 fast5 files in Fast5_20180131_1523_7fosmid_RLB001-pass/
Classifying fast5s: 25600 / 25600 (100.0%)
Moving fast5s:      25600 / 25600 (100.0%)

Barcode     Count
      1        33
      2        98
      3      5818
      4      2010
      5      1076
      6      1985
      7      1091
      8      1964
      9      1074
     10       364
     11       388
     12        94
   none      9605


Found 80,557 fast5 files in Fast5_20180131_1523_7fosmid_RLB001-pass/
Classifying fast5s: 25600 / 25600 (100.0%)
Moving fast5s:      25600 / 25600 (100.0%)

Barcode     Count
      1        42
      2       111
      3      5898
      4      2041
      5      1073
      6      1963
      7      1017
      8      1984
      9      1110
     10       278
     11       391
     12        99
   none      9593