tomodrgn graph_traversal

Purpose

Identify particle indices forming the most efficient path through latent space, while connecting specified starting and ending points by way of specified anchor points.

Sample usage

The examples below are adapted from tomodrgn/testing/commandtest*.py, and rely on other outputs from commandtest.py to execute successfully.

# Warp v1 style inputs
tomodrgn \
    graph_traversal \
    output/vae_both_sim_zdim8_dosetiltweightmask_batchsize8/z.39.train.pkl \
    --anchors 5 10 15 20 \
    -o output/vae_both_sim_zdim8_dosetiltweightmask_batchsize8/graph_traversal \
    --max-neighbors 20 \
    --avg-neighbors 20

# WarpTools style inputs
tomodrgn \
    graph_traversal \
    output/vae_warptools_70S_zdim8_dosetiltweightmask_batchsize8/z.39.train.pkl \
    --anchors 5 10 15 20 \
    -o output/vae_warptools_70S_zdim8_dosetiltweightmask_batchsize8/graph_traversal \
    --max-neighbors 20 \
    --avg-neighbors 20

Arguments

usage: graph_traversal [-h] --anchors ANCHORS [ANCHORS ...] -o OUTDIR
                       [--max-neighbors MAX_NEIGHBORS]
                       [--avg-neighbors AVG_NEIGHBORS]
                       [--plot-format {png,svgz}]
                       z

Positional Arguments

z

Input latent embeddings z.pkl file

Core arguments

--anchors

Indices of anchor points along desired trajectory. At least 2 points must be specified.

-o, --outdir

Directory in which to store output .txt/.pkl files of path indices and latent embeddings

--max-neighbors

The maximum number of neighbors to initially calculate distances for from each latent embedding

Default: 10

--avg-neighbors

Used to set a cutoff distance defining connected neighbors such that each embedding will have this many connected neighbors on average

--plot-format

Possible choices: png, svgz

File format with which to save plots

Default: 'png'

Common next steps

• Validate the inferred latent space graph traversal by isolating indices of particles proximal to each neighbor point or anchor point along the path, and performing homogeneous reconstructions with tomodrgn backproject_voxel or external STA software