Oh, and if you have existing data with DPC barcodes - WDX and its models (including one trained on DPC) are available here:

github.com/KleistLab/Wa...

I won't go into too much detail on the rational barcode design (check it out in Fig 2), but briefly, we searched the DNA signal space for sets of sequences with high DTW distances between them.

We picked a set of 12 bc's to establish WDX, but this is certainly not the limit ...

As such, Wiep van der Toorn implemented a signal segmentation logic that reduced the required DTW computations by orders of magnitude. As added benefit, this step also reduces the negative impact of longer motor protein stalls.

Now, with a working classifier, we looked at barcode design.

So instead of basecalling, we decided to use dynamic time warping to classify signals. DTW is commonly used for nanopore data, because it is tolerant of variation in the timing of motor protein steps.

However, calculating it on raw data quickly becomes computational extensive.

A brief summary: We were inspired by DeePlexiCon's strategy of altering the DNA adapter ligated onto the RNA. This doesn't add complexity to library prep, DNA oligos cost a lot less than RNA. However, there's a big problem: the DNA signal in dRNA seq cannot be basecalled.

This was one of my last projects at @Helmholtz_HIRI
and an incredible fun collaboration with Max van der Kleist's team at RKI Berlin @rki_de
- especially with Wiep, who's the mastermind behind the barcode design and classification algorithms that are at the core of WarpDemuX.