circuit-rs 0.4.0

A library for doing efficient operations on neural circuits

rusted_circuits

Exposing useful operations to manipulate the neural circuits that can be derived from Flywire, a project which maps out the Drosophila brain.

Context

In my research, I found myself repeatedly needing to find a minimal circuit (connectivity-wise) from a given set of source and sink nodes. Specifically, I needed to identify the smallest subcircuit containing all nodes on all simple paths between a set of (source, sink) nodes. The python code I used was too slow, even considering various biological priors (e.g. I think it's possible to traverse from most neurons to almost any other within 5 jumps, and if you can't then it's probably not possible), and existing solutions such as networkx-all_simple_paths, and rustworkx-all_simple_paths only consider single source-sink pairs at a time, so I'd need rerun the algorithm #Source * #Sink times, but by writing my own script, I only needed to run it #Source times. Also, those libraries return lists-of-lists, which can cause lots of memory to build up and since I only care about the nodes and not the actual path, I figure I can avoid that expense.

Design

We accept numpy arrays of 2D where column 1 is the pre-synaptic and column 2 is the post-synaptic.

Functions

isolate_subcircuit(
    edge_array: np.ndarray[u64, u64]
    sources: set[int],
    sinks: set[int],
    max_depth: int,
    max_paths: int,
    nid_to_cidx_map: dict[int,int] | None
  
)
  
map_neuron_ids_to_cmat_idxs(
  edge_array: np.ndarray[u64, u64],
)

adjacency_map_from_edge_list(
  edge_array=np.ndarray[u64, u64],
  nid_to_cidx_map=dict[int, int] | None
)

Example

Smallest Subcircuit

Here is a simple working example, where we have a linear path from source (1) to sink (5)

edge_list = np.asarray([(1, 2), (2, 3), (3, 4), (4, 5)])
sources = {1}
sinks = {5}
max_depth = 10
max_paths = 100

# The nid_to_cidx_map is optional, and if it is not specified, we derive the mapping of the neuron-id-to-connectivity-matrix-index.
nid_to_cidx_map = {1: 10, 2: 20, 3: 30, 4: 40, 5: 50}

# Expected: all nodes on the path from source to sink
expected = {10, 20, 30, 40, 50}
actual_np = isolate_subcircuit_np_edgelist(edge_list, sources, sinks, max_depth, max_paths, nid_to_cidx_map)

# -OR-, from a list
actual_list = isolate_subcircuit(edge_list.tolist(), sources, sinks, max_depth, max_paths, nid_to_cidx_map)

Maintainers

1. Add code/documentation changes via git add ...

2. Attach a commit message e.g. git commit -m '...'

Note: in step 3, replace VERSION_NUMBER below by an actual version number e.g. v0.1.0

3. git tag $VERSION_NUMBER

4. Modify Cargo.toml, the version to be the same (not strictly necessary, but it's just nicer).

5. git push origin $VERSION_NUMBER