inconsistency converting between Flycircuit and Hemibrain

[George Yerou]

Dear Dr Jefferis/ NBLAST community,

Based on colourdepth mask search, the most similar neurons to VGlut-F-300133 include bodyids  1040013335 (descending neuron) and 1067240551 (CL37). However, using swcs of these EM neurons as inputs in nblast (after conversion to FCWB template), VGlut-F-300133 is not among the top 10 outputs. For 1040013335 the top results is Cha-F-100034 and for 1067240551  the top results look very different, with the top scorers being TH-F-200020 and TH-M-000060. I understand that the flycircuit dataset is incomplete but surley the closest match for a flycircuit neuron in the EM dataset should also be the same when the direction is reversed?

NB: 

• I am using the "mirror" and "compare with all neurons" settings when uploading EM SWCs to NBLAST.
• to colourdepth mask search EM neurons i am converting the SWC to the colordepth jrcu2018 template, dilating and using the psychadelic rainbow lut technique.

Best wishes,

George.

[Greg Jefferis]

Are you also using the "Use mean scores" option? NBLAST is asymmetric. The score for a big neuron vs a small neuron may be low even if the score for a small neuron vs a big neuron is high. See the NBLAST paper for more details of this. TH-XXX neurons are likely very big with arbours across large swathes of the brain. Note that MIP search and NBLAST do different things so that it is not guaranteed that you will get the same answer. Also you should make sure to process your EM neurons by pruning twigs less than 2-5µm before making the dot props object. Best, Greg.