Running NBLAST
Scoring morphological similarity yourself, in navis or nat.nblast, rather than using VFB’s precomputed scores.
VFB precomputes NBLAST scores and exposes them through the “find similar” queries on the website, so you do not need to run NBLAST yourself to use them — the NBLAST concept page explains what is covered and how to tell whether a given neuron has scores. This page is for when you want to score your own neurons, or score against a set VFB does not cover.
NBLAST works on dotprops: neurons resampled into points with associated tangent vectors.
Scoring is therefore a two-step business — convert, then compare — and both the resampling
distance and the neighbourhood size affect the result. Neurons must also be in the same
template space before the comparison means anything, which is where
flybrains or nat.flybrains come in.
In Python, with navis
import navis
nl = navis.example_neurons(5)
# Convert to dotprops first; resample and k affect the scores
dps = navis.make_dotprops(nl, resample=1, k=5)
scores = navis.nblast_allbyall(dps)
navis.nblast() compares one set against another where you do not want the all-by-all matrix.
Scores are conventionally normalised against a self-self comparison, so a perfect match scores
1.
In R, with nat.nblast
nat.nblast is part of the natverse and installs with it.
Where next
The NBLAST tutorial works through a real comparison end to end, including preparing the dotprops and interpreting the resulting scores. For what the scores mean and how VFB uses them, see NBLAST.
The method is described in Costa et al. (2016).