tomodrgn eval_vol#

Purpose#

Generate volumes from corresponding latent embeddings using a pretrained train_vae model (i.e. evaluating decoder module only).

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 \
    eval_vol \
    --weights output/vae_both_sim_zdim2/weights.pkl \
    -c output/vae_both_sim_zdim2/config.pkl \
    -o output/vae_both_sim_zdim2/eval_vol_allz \
    --zfile output/vae_both_sim_zdim2/z.train.pkl \
    -b 32

# WarpTools style inputs
tomodrgn \
    eval_vol \
    --weights output/vae_warptools_70S_zdim2/weights.pkl \
    -c output/vae_warptools_70S_zdim2/config.pkl \
    -o output/vae_warptools_70S_zdim2/eval_vol_allz \
    --zfile output/vae_warptools_70S_zdim2/z.train.pkl \
    -b 32

Arguments#

usage: eval_vol [-h] -w WEIGHTS -c CONFIG -o OUTDIR [--prefix PREFIX]
                [--zfile ZFILE] [--flip] [--invert] [--downsample DOWNSAMPLE]
                [--lowpass LOWPASS] [-b BATCH_SIZE] [--no-amp] [--multigpu]

Core arguments#

-w, --weights

Model weights from train_vae

-c, --config

config.pkl file from train_vae

-o, --outdir

Output .mrc or directory

--prefix

Prefix when writing out multiple .mrc files

Default: 'vol_'

Specify z values#

--zfile

Text/.pkl file with z-values to evaluate

Volume arguments#

--flip

Flip handedness of output volume

Default: False

--invert

Invert contrast of output volume

Default: False

--downsample

Downsample volumes to this box size (pixels)

--lowpass

Lowpass filter to this resolution in Å. Requires settings –Apix.

Compute arguments#

-b, --batch-size

Batch size to parallelize volume generation (32-64 works well for box64 volumes)

Default: 32

--no-amp

Disable use of mixed-precision training

Default: False

--multigpu

Parallelize model evaluation across all detected GPUs. Specify GPUs i,j via export CUDA_VISIBLE_DEVICES=i,j

Default: False

Common next steps#

  • Analyze a volume ensemble using dimensionality reduction with tomodrgn analyze_volumes

  • Use external tools such as MAVEn or SIREn to systematically quantify structural heterogeneity either with or without an atomic model to guide analysis