NBLAST

What is NBLAST?

NBLAST (Costa et al., 2016) is a computational method to quantify morphological similarity between neurons. It provides an objective way to compare neuron shapes and identify morphologically similar cells within and across datasets.

How NBLAST works

NBLAST operates on “dotprops” - a representation of neurons as tangent vectors that capture the local geometry of neuronal arbors. The algorithm:

Converts neurons to dotprops: Each neuron is represented as a set of points with associated directional vectors
Compares vector pairs: For each tangent vector in a query neuron, NBLAST finds the closest tangent vector in the target neuron
Calculates similarity scores: Scores are computed based on both the distance between vectors and their directional similarity (dot product)
Normalizes results: Final scores are typically normalized to a self-self comparison, where a perfect match equals 1

NBLAST concept

NBLAST on VFB

VFB precomputes NBLAST scores so that morphological similarity searches are fast and need no computational expertise. Coverage is not uniform: scores exist where a neuron has a registered morphology in a template space that has been scored, and they are added as new data lands. The reliable way to tell whether a given neuron has them is to open its Term Info — the similarity query appears only when scores exist for that neuron.

What’s included

Scores cover individual neurons from the EM connectome datasets listed on the EM data page, single neurons from the FlyCircuit collection, and split-GAL4 expression patterns from FlyLight — the last of these being what lets you go from a neuron to a driver line that might label it.

The most recent expansion of coverage is described in the precomputed similarity scores announcement.

Types of comparisons

Neuron-to-neuron: Find morphologically similar neurons within or across datasets
Neuron-to-expression pattern: Identify split-GAL4 lines that might label neurons similar to those in connectome datasets

Using NBLAST on VFB

Accessing NBLAST queries

See Similarity Score Queries Guide for details of how to find NBLAST similarity queries in the VFB interface.

Interpreting results

Scores on VFB are normalised against a self-self comparison, so 1.0 is the score a neuron gets against itself and is the practical ceiling. Higher is more similar; negative scores indicate morphologies with little in common.

There is no universal threshold that separates a real match from a spurious one. The cut-off depends on the neuron, the datasets being compared and the quality of the reconstructions, so a score should be read against the other scores in the same ranked list rather than against a fixed scale. A hit at the top of a list of otherwise low scores is worth inspecting; the same absolute value buried in a list of high scores is not. Always look at the morphology before accepting a match.

Applications

Research applications

Cell type classification: Group neurons by morphological similarity
Cross-dataset comparison: Find corresponding cell types across different connectomes
Driver line selection: Identify genetic tools for targeting specific morphological types
Evolutionary studies: Compare homologous neurons across species

Workflow integration

NBLAST results on VFB can be:

Exported for further analysis
Used to build custom neuron collections
Combined with other search criteria (anatomy, connectivity)
Accessed programmatically via VFB APIs

Technical considerations

Optimization for VFB

All neurons are standardized to common template spaces
Consistent spatial resolution across datasets
Normalized scoring for cross-dataset comparisons
Regular updates as new data becomes available

Limitations

Focuses purely on morphological similarity
May not capture functional relationships
Sensitive to differences in reconstruction quality
Template registration accuracy affects cross-dataset comparisons