Conversation
|
@eliottpark do you want to have a quick check and lmk? I don't have more time today, but will run my tests and check if this has broken other stuff in the next few days. |
|
Tested all combinations (VH + linker + VL, VH + linker + VH, VL + linker + VL, VL + linker + VH) with a few different sequences on accuracy mode and the model identified all variable regions accurately with 0 linkers in between. Thank you for the rapid support! |
|
Hey @ALGW71 , one more thing I just came across - happy to spin this into a separate issue or question. Is there a practical limit on the sequence length that you'd suggest? For a sequence containing two identical VHs, I've observed that by concatenating Alanines to the end of the sequence, the variable region predictions start to become unstable. I didn't get to a magic number, but for some examples it seemed that under length 400, both VHs are identified correctly. After appending "A"s to get to a total sequence length of over 400, the second VH is not identified. This does seem to be dependent on sequence composition - I tried modifying my VH and CH1 independently and both seem to have an impact on the success of the prediction. For reference, in one case, modifying a VH within a VH + CH1 + G4Sx4 + VH + CH1 construct with a single residue substitution resulted in a successful prediction while the WT only identified the first VH. |
2 participants
Solves #116
Not tested.