Hi swim1128,

Remora indeed cannot handle ambiguous bases in motifs directly during chunk dataset preparation, as its reference scanning expects unambiguous sequences. A straightforward workaround is to expand your degenerate motif (GTNNaNNTGG) into all possible specific 10-mers by substituting each N (and assuming the lowercase 'a' denotes A) with A, C, G, or T. This generates 256 variants for the four N positions. You can then list them all as a comma-separated string in the --motifs argument when running remora prepare_chunks on your reference. The tool supports multiple motifs this way, allowing it to extract relevant signal chunks across the expanded set without issue. Once prepared, proceed with training as usual.

Kevin