Examples using API

graynet 2.0 keeps the classic entrypoints for computing outputs, but the result layout is now run-level and canonical:

• run_metadata.json

• edges_raw.parquet

• edges_summary.parquet when summary statistics are requested

• roi_stats.parquet when ROI statistics are requested

Single-feature edge extraction

from pathlib import Path
from graynet import extract

input_dir = Path("/project/freesurfer_reconall")
subjects = ["sub-001", "sub-002"]

run_dir = extract(
    subjects,
    input_dir,
    base_feature="freesurfer_thickness",
    weight_method_list=["manhattan", "cosine"],
    atlas="fsaverage",
    smoothing_param=10,
    out_dir=Path("/project/graynet_runs"),
    return_results=False,
    num_procs=2,
)

print(run_dir)

Reading run outputs

from graynet import load_run, get_edge_values, export_to_nx

run = load_run(run_dir)
print(run.metadata["subject_ids"])
print(run.metadata["node_labels"][:5])

edge_rows = get_edge_values(
    run.raw_edges,
    subject_id="sub-001",
    base_feature="freesurfer_thickness",
    weight_method="manhattan",
)

X = run.raw_edges.to_ndarray(
    ["sub-001", "sub-002"],
    base_feature="freesurfer_thickness",
    weight_method="manhattan",
)

graph = export_to_nx(edge_rows)

Multi-edge extraction

from graynet import extract_multiedge

run_dir = extract_multiedge(
    subjects,
    input_dir,
    base_feature_list=["freesurfer_thickness", "freesurfer_curv"],
    weight_method_list=["manhattan"],
    summary_stats=["median", "prod"],
    atlas="fsaverage",
    smoothing_param=10,
    out_dir=Path("/project/graynet_runs"),
    return_results=False,
)

ROI-wise statistics

import numpy as np
from graynet import roiwise_stats_indiv

run_dir = roiwise_stats_indiv(
    subjects,
    input_dir,
    base_feature="freesurfer_thickness",
    chosen_roi_stats=["median", np.nanmean],
    atlas="fsaverage",
    smoothing_param=10,
    out_dir=Path("/project/graynet_runs"),
    return_results=False,
)