TomoTwin is an application the enables particle picking for Cryo-ET using deep metric learning based procedures.

TomoTwin comes pre-trained on so far 120 different proteins. By embedding tomograms in an information-rich, high-dimensional space which separates macromolecules according to their 3-dimensional structure, TomoTwin allows users to identify proteins in tomograms de novo without manually creating training data or retraining the network each time a new protein is to be located. That means, you can simply run it for your specific sample without a much additional effort.

Reference:

• Rice, G., Wagner, T., Stabrin, M. et al. TomoTwin: generalized 3D localization of macromolecules in cryo-electron tomograms with structural data mining Nat Methods (2023)

• TomoTwin Site
• TomoTwin Github

This application uses the napari application to visualize the tomograms. napari requires a graphical connection using NX

• Module Name: tomotwin (see the modules page for more information)
• GPU enabled
• Environment variables set
  • TOMOTWIN_TEST_DATA
  • TOMOTWIN_MODEL
• Example files in $TOMOTWIN_TEST_DATA
• Model file in $TOMOTWIN_MODEL

Allocate an interactive session and run the program.
Sample session (user input in bold):

[user@biowulf]$ sinteractive --mem=16G --gres=lscratch:50,gpu:v100x:2
salloc.exe: Pending job allocation 46116226
salloc.exe: job 46116226 queued and waiting for resources
salloc.exe: job 46116226 has been allocated resources
salloc.exe: Granted job allocation 46116226
salloc.exe: Waiting for resource configuration
salloc.exe: Nodes cn3144 are ready for job

[user@cn3144 ~]$ module load tomotwin

[user@cn3144 ~]$ cd /lscratch/$SLURM_JOB_ID

[user@cn3144 46116226]$ cp -r $TOMOTWIN_TEST_DATA/* .

[user@cn3144 46116226]$ CUDA_VISIBLE_DEVICES=0,1 tomotwin_embed.py tomogram \
  -m tomotwin_model_p120_052022_loss.pth \
  -v tomo/tomo.mrc \
  -o out/embed/tomo/ \
  -b 400
Latest version of TomoTwin is installed :-)
reading tomotwin_model_p120_052022_loss.pth
Model config:
{'identifier': 'SiameseNet', 'network_config': ...}
... UserWarning: This DataLoader will create 12 worker processes in total. Our suggested max number of worker in current system is 4 ...
Embeddings have shape: (5083356, 35)
Wrote embeddings to disk to out/embed/tomo/tomo_embeddings.temb
Done.

[user@cn3144 46116226]$ tomotwin_tools.py extractref \
  --tomo tomo/tomo.mrc \
  --coords ref.coords \
  --out out/extracted_ref/
1it [00:00, 97.91it/s]
wrote subvolume reference to out/extracted_ref/

[user@cn3144 46116226]$ CUDA_VISIBLE_DEVICES=0,1 tomotwin_embed.py subvolumes \
  -m tomotwin_model_p120_052022_loss.pth \
  -v out/extracted_ref/reference_0.mrc \
  -o out/embed/ref
Latest version of TomoTwin is installed :-)
reading tomotwin_model_p120_052022_loss.pth
Model config:
{'identifier': 'SiameseNet', 'network_config': ...}
UserWarning: This DataLoader will create 12 worker processes in total. Our suggested max number of worker in current system is 4 ...
Done. Wrote results to out/embed/ref/embeddings.temb

[user@cn3144 46116226]$ tomotwin_map.py distance \
  -r out/embed/ref/embeddings.temb \
  -v out/embed/tomo/tomo_embeddings.temb \
  -o out/map/
Latest version of TomoTwin is installed :-)
Read embeddings
Map references: 100%|████████████████████████████████████| 1/1 [00:04<00:00,  4.55s/it]
Prepare output...
Wrote output to out/map/map.tmap

[user@cn3144 46116226]$ tomotwin_locate.py findmax -m out/map/map.tmap -o out/locate/
Latest version of TomoTwin is installed :-)
start locate  reference_0.mrc
effective global min: 0.5
Locate class 0: 100%|███████████████████████████████| 31071/31071 [00:06<00:00, 4604.62it/s]
Call get_avg_pos
done 0
Located reference_0.mrc 847
Non-maximum-supression: 100%|███████████████████████| 847/847 [00:00<00:00, 3489.54it/s]
Particles of class reference_0.mrc: 844 (before NMS: 847)

[user@cn3144 46116226]$ exit
salloc.exe: Relinquishing job allocation 46116226
[user@biowulf ~]$

Create a batch input file (e.g. tomotwin.sh). For example:

#!/bin/bash
set -e
cd /lscratch/$SLURM_JOB_ID
module load tomotwin
cp -r $TOMOTWIN_TEST_DATA/* .

CUDA_VISIBLE_DEVICES=0,1 tomotwin_embed.py tomogram \
    -m tomotwin_model_p120_052022_loss.pth \
    -v tomo/tomo.mrc \
    -o out/embed/tomo/ \
    -b 400
tomotwin_tools.py extractref \
    --tomo tomo/tomo.mrc \
    --coords ref.coords \
    --out out/extracted_ref/
CUDA_VISIBLE_DEVICES=0,1 tomotwin_embed.py subvolumes \
    -m tomotwin_model_p120_052022_loss.pth \
    -v out/extracted_ref/reference_0.mrc \
    -o out/embed/ref
tomotwin_map.py distance \
    -r out/embed/ref/embeddings.temb \
    -v out/embed/tomo/tomo_embeddings.temb \
    -o out/map/
tomotwin_locate.py findmax -m out/map/map.tmap -o out/locate/

Submit this job using the Slurm sbatch command.

sbatch [--cpus-per-task=#] [--mem=#] [--gres=lscratch:#,gpu:type:2] tomotwin.sh

Create a swarmfile (e.g. tomotwin.swarm). For example:

CUDA_VISIBLE_DEVICES=0,1 tomotwin_embed.py tomogram -m model.pth -v tomo1.mrc -o out1 -b 400
CUDA_VISIBLE_DEVICES=0,1 tomotwin_embed.py tomogram -m model.pth -v tomo2.mrc -o out2 -b 400
CUDA_VISIBLE_DEVICES=0,1 tomotwin_embed.py tomogram -m model.pth -v tomo3.mrc -o out3 -b 400
CUDA_VISIBLE_DEVICES=0,1 tomotwin_embed.py tomogram -m model.pth -v tomo4.mrc -o out4 -b 400

Submit this job using the swarm command.

swarm -f tomotwin.swarm [-g #] [--gres=gpu:type:2] --module tomotwin