This is the multi-page printable view of this section. Click here to print.
Concepts
- 1: Transgene Expression
- 2: Split Expression
- 3: NBLAST
- 4: Templates
- 5: Bridging Registrations
- 6: Cell Types
- 7: Confidence Values
- 8: FlyLight Adult CNS Imaging Tiles
1 - Transgene Expression
Here we present details of how single transgenes are curated

Expression of single transgenes are curated from the published literature into FlyBase as expression statements. Virtual Fly Brain (VFB) combines hosted images with neuroanatomical, expression and genetic data from FlyBase in Neo4j and maps them onto Central Nervious System (CNS) templates where they can be queried using Web Ontology Language (OWL) and Solr.
2 - Split Expression
Thousands of hemi-driver transgenes are now available, meaning that millions of combinations/splits are possible, each targeting some precise subset of the 10s-100s of thousands of neurons in a fly Central Nervous System (CNS). Finding the right combination for an experiment is a serious bottleneck for researchers.
FlyBase & Virtual Fly Brain (VFB) solve this problem by curating information and images recording where expression is driven by combinations of hemi-drivers. Each combination gets a standard name on VFB reflecting the names of the component hemidrivers (see figure below) and is also associated with any short names for the combination used in the literature (e.g. “P{VT017411-GAL4.DBD} ∩ P{VT019018-p65.AD} expression pattern” has_exact_synonym: SS02256")
We use programmatic methods to generate expression statements for hemi-driver combinations with FlyLight image data hosted by VFB. Combinations are validated against a local copy of the FlyLight dataset and a structured user readable comment is generated as part of the expression statement. These hyperlink to the partner hemi-driver in FlyBase, contain the strain designation (where available) and can be automatically parsed by Virtual Fly Brain to attach expression and genetic data to a node representing the intersection.

3 - 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 on VFB
VFB provides precomputed NBLAST scores for all neurons in its database, making morphological similarity searches fast and accessible without requiring computational expertise.
What’s included
VFB NBLAST scores cover:
-
All individual neurons from major connectome datasets including:
- FAFB-FlyWire
- Male-CNS optic lobe
- Hemibrain
- FlyCircuit
- FAFB-CATMAID datasets
-
Split-GAL4 expression patterns from FlyLight, enabling discovery of potential driver lines that label specific morphological types
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
NBLAST similarity searches are available directly in the VFB interface:
- Navigate to any individual neuron or split-GAL4 image page
- Scroll to the term info panel
- Look for NBLAST query options (only appears if results are available)
- Click to see ranked lists of morphologically similar items
Interpreting results
NBLAST scores range from -1 to 1:
- 1.0: Perfect match (identical morphology)
- 0.5-1.0: High similarity
- 0.0-0.5: Moderate similarity
- Below 0.0: Low similarity or dissimilar morphologies
Results are typically ranked by score, with the most similar items appearing first.
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
Further reading
- NBLAST tutorial - Detailed programming tutorial
- Original NBLAST paper - Costa et al., 2016
- VFB NBLAST announcement - Recent updates and expanded coverage
4 - Templates
Many central nervious system (CNS) templates exist for Drosophila, below we provide a summary of those used in VFB.
Central Nervious System from Janelia Research Campus JRC2018
VFB displays data aligned to either the JRC 2018 unisex brain template for cephalic or JRC 2018 unisex Ventral Nerve Cord (VNC) for non-cephalic data.
Bogovic et al., “An unbiased template of the Drosophila brain and ventral nerve cord”
We recomend downloading templates from the VFB browser or the links below however the original files are available from Janelia: JRC 2018 Brain templates
Adult Brain from Janelia Research Campus/VFB (JFRC2010/JFRC2)
Superceeded by JRC2018

Template brain created by Arnim Jenett (Janelia Research Campus), Kazunori Shinomiya and Kei Ito (Tokyo University) from a staining with the neuropil marker nc82.
The voxel size is 0.62x0.62x0.62 micron.
The files are available here.
Painted domains/regions in the available templates
5 - Bridging Registrations
Many transforms to map between different Drosophila template brains are available.
You can see how to use the above in python using navis-flybrains or in R using nat.flybrains
6 - Cell Types
Neurons on VFB are annotated with cell types from the Drosophila Anatomy Ontology (FBbt).

Why do we use ontology terms?
- Each term represents a concept of a cell type, with a definition based on referenced publications:
- As well as a label, each term has a collection of synonyms, facilitating identification even when the same type has been referred to by different names in different sources:
- Hierarchical – e.g. specific terms for MBON01, MBON02 etc., but also grouped by a general MBON term and all under ‘adult neuron’
- Neurons of the same type in multiple datasets can be linked to the same ontology term
- Persistent, resolvable identifiers to uniquely identify cell types e.g. http://virtualflybrain.org/reports/FBbt_00100234
We also use terms from the Drosophila Anatomy Ontology to annotate CNS regions (for the Template ROI Browser
tool and neuron connectivity per region
query) and other anatomical features.
7 - Confidence Values
Some annotations on VFB are based on predictions, for example, predicted neurotransmitters for neurons in electron microscopy datasets (see below). Where available, we include the confidence of these predictions (as a badge next to the annotation), as well as a link to the publication - in the example below, this would be by clicking on the ‘DOI’ badge.

Neurotransmitter Prediction Confidence Values
Neurotransmitter predictions for neurons in the FlyWire FAFB and Hemibrain datasets are the ‘conf_nt’ predictions from Eckstein et al. (2024). MANC and optic-lobe predictions are from Takemura et al. (2023) and Nern et al. (2024), respectively, using the methodology of Eckstein et al. (2024). This analysis only investigated a limited set of neurotransmitters and assumed a single neurotransmitter per neuron (see publications for further detail). Neurons with fewer than 100 presynapses are excluded.
8 - FlyLight Adult CNS Imaging Tiles
While most first-generation FlyLight expression pattern images are single 20x objective acquisitions, many Split-GAL4 lines were imaged using 63x or 40x objectives. This requires multiple image acquisitions to cover the CNS regions of interest. VFB integrates only the combined images derived from these tiles and lists the tiles used to create these combined images in the comment section for each image, for example: “tile(s): ’left_dorsal, right_dorsal, ventral”. These tiles do not necessarily cover the entire brain or VNC, which complicates the assessment of the specificity of these lines. VFB notes in the comment where the combined tiles do not provide full coverage. For further information refer to Figure 4 Supplemental File 1 from ‘A split-GAL4 driver line resource for Drosophila CNS cell types’ direct link. This is a preprint and may be updated.
Sets of tiles from FlyLight (Meissner et al., 2024):




