fafbseg.flywire.synapses.get_transmitter_predictions

fafbseg.flywire.synapses.get_transmitter_predictions(x, single_pred=False, weighted=True, materialization='auto', filtered=True, neuropils=None, batch_size=10, *, dataset=None, **kwargs)[source]

Fetch neurotransmitter predictions for neurons.

Based on Eckstein et al. (2020). The per-synapse predictions are collapsed into per-neuron prediction by calculating the average confidence for each neurotransmitter across all synapses weighted by the “cleft score”. Bottom line: higher confidence synapses have more weight than low confidence synapses.

Parameters:

  • x (int | list of int | Neuron/List | NeuronCriteria) – Either a FlyWire segment ID (i.e. root ID), a list thereof or a Neuron/List. For neurons, the .id is assumed to be the root ID. If you have a neuron (in FlyWire space) but don’t know its ID, use fafbseg.flywire.neuron_to_segments() first.

  • single_pred (bool) – Whether to only return the highest probability transmitter for each neuron.

  • weighted (bool) – If True, will weight predictions based on confidence: higher cleft score = more weight.

  • materialization (int | str, optional) –

    Which materialization to query:

    • ’auto’ (default) tries to find the most recent materialization version at which all the query IDs existed

    • ’latest’ uses the latest materialized table

    • ’live’ queries against the live data - this will be much slower!

    • pass an integer (e.g. 447) to use a specific materialization version

  • filtered (bool) – Whether to use the filtered synapse table. Briefly, this filter removes redundant and low confidence (<= 50 cleft score) synapses. See also https://tinyurl.com/4j9v7t86 (links to CAVE website).

  • neuropils (str | list of str, optional) – Provide neuropil (e.g. 'AL_R') or list thereof (e.g. ['AL_R', 'AL_L']) to filter predictions to these ROIs. Prefix neuropil with a tilde (e.g. ~AL_R) to exclude it.

  • batch_size (int) – Number of IDs to query per batch. Too large batches might lead to truncated tables: currently individual queries can not return more than 200_000 rows and you will see a warning if that limit is exceeded.

  • dataset ("public" | "production" | "sandbox", optional) – Against which FlyWire dataset to query. If None will fall back to the default dataset (see set_default_dataset()).

  • **kwargs – Keyword arguments are passed through to fafbseg.flywire.get_synapses().

Returns:

  • pandas.DataFrame – If single_pred=False: returns a dataframe with all per-transmitter confidences for each query neuron.

  • dict – If single_pred=True: returns dictionary with (top_transmitter, confidence) named tuple for each query neuron.

Examples

>>> from fafbseg import flywire

Get per-transmitter predictions for a single neuron:

>>> flywire.get_transmitter_predictions(720575940603231916)
Using materialization version 783.
root_id        720575940603231916
gaba                     0.011677
acetylcholine            0.938961
glutamate                0.017902
octopamine               0.012861
serotonin                0.012467
dopamine                 0.006132

Return only the most likely transmitter:

>>> flywire.get_transmitter_predictions(720575940603231916, single_pred=True)
Using materialization version 783.
{720575940603231916: prediction(transmitter='acetylcholine', probability=0.9389612897479809)}