DeepMM: full-length protein structure determination from cryo-EM maps using deep learning.

DeepMM: full-length protein structure determination
from cryo-EM maps using deep learning.

Quick Links

DeepMM implements fully automated de novo structure modeling method, MAINMAST, which builds three-dimensional models of a protein from a near-atomic resolution EM map. The method directly traces the protein’s main-chain and identifies Cα positions as tree-graph structures in the EM map.

DeepMM uses MAINMAST to trace the main chain paths on main chain probability maps.

References:

Terashi, Genki, and Daisuke Kihara
De novo main-chain modeling for EM maps using MAINMAST.
Nature communications (2018), 9(1), 1618.
Terashi, Genki, and Daisuke Kihara
De novo main-chain modeling with MAINMAST in 2015/2016 EM Model Challenge.
Journal of Structural Biology (2018), 204(2), 351-359.

Documentation

DeepMM Github page

Important Notes

Module Name: DeepMM (see the modules page for more information)
Unusual environment variables set
- DEEPMM_HOME installation directory
- DEEPMM_BIN executable directory
- DEEPMM_SRC a folder containing the source code
- DEEPMM_DATA a folder containing sample data

Interactive job

Interactive jobs should be used for debugging, graphics, or applications that cannot be run as batch jobs.

Before running an interactive session, install Jackal software locally in your account:
1) following the link:
https://honiglab.c2b2.columbia.edu/software/cgi-bin/software.pl?input=Jackal
manually download the file jackal_64bit.tar.gz to your local computer
2) sftp/transfer the file to Biowulf
3) ungzip/untar it:
tar -zxf jackal_64bit.tar.gz
and
4) move the resulting folder jackal_64bit
to the directory: /data/$USER

Allocate an interactive session and run the program. Sample session:

[user@biowulf]$ sinteractive --mem=8g  --gres=gpu:k80:1,lscratch:10 -c4 
[user@cn4213 ~]$ module load DeepMM 
[+] Loading CUDA Toolkit  10.2.89  ...
[+] Loading cuDNN/7.6.5/CUDA-10.2 libraries...
[+] Loading blast 2.11.0+  ...
[+] Loading DeepMM  20210722

Set the environment variables:

[user@cn4213 ~]$  export JACKALDIR=/data/$USER/jackal_64bit
[user@cn4213 ~]$  PATH=/data/$USER/jackal_64bit/bin:$PATH 
[user@cn4213 ~]$  git clone https://github.com/JiahuaHe/DeepMM

Run DeepMM on the sample data:

[user@cn4213 ~]$ cd DeepMM/5185
[user@cn4213 ~]$ preprocess.py -m 5185_zoned.mrc -n map.npz -t 3.21  
[INFO] 10733 voxels with density value greater than 3.21000 are retained.
[INFO] 10733 voxels are saved to map.npz

[user@cn4213 ~]$ pred_MCCA.py -n map.npz -m predMC.mrc -c predCA.mrc 
LOAD] loading data...
[PREDICT] predicting...
100%|████████████████████████████████████████████████████████████████████████████████████████████████| 84/84 [00:04<00:00, 17.43it/s]

[user@cn4213 ~]$  mrc2situs.py -m 5185_zoned.mrc -s map.situs 
[user@cn4213 ~]$  mrc2situs.py -m predMC.mrc -s predMC.situs    
[user@cn4213 ~]$  mrc2situs.py -m predCA.mrc -s predCA.situs    

[user@cn4213 ~]$  getldp predMC.situs predCA.situs > LDP.pdb    

[user@cn4213 ~]$  getvox map.situs LDP.pdb > LDP.mcv 

[user@cn4213 ~]$  pred_AASS.py -m LDP.mcv -l LDP.pdb > LDP_AASS.pdb 
100%|█████████████████████████████████████████████████████████████████████| 7/7 [00:03<00:00,  2.18it/s]

[user@cn4213 ~]$  trace LDP_AASS.pdb -nrd 100 > paths.pdb 

[user@cn4213 ~]$  align paths.pdb seq.fasta.spd3 
# ALIGN.F
 # Align sequence to main-chain paths
 # Parameter setting:
 # wca =   1.600000
 # whelix =   1.000000
 # wsheet =  0.7000000
 # wcoil =  0.8000000
 # waa =   1.000000
 # wss =  0.5000000
 # rmsd =   5.000000
 # nout =          10
 # prefix = model
 # Reading LDP paths from file paths.pdb
 # Path            1  consists of          565  LDPs.
 # Path            2  consists of          560  LDPs.
 # Path            3  consists of          534  LDPs.
 # Path            4  consists of          538  LDPs.
 # Path            5  consists of          511  LDPs.
 # Path            6  consists of          520  LDPs.
 # Path            7  consists of          543  LDPs.
 # Path            8  consists of          484  LDPs.
 # Path            9  consists of          551  LDPs.
 # Path           10  consists of          475  LDPs.
 # Number of paths:           10
 # Reading SS prediction from file seq.fasta.spd3
 # Reading predicted SS from .SPD3 file seq.fasta.spd3
 # Number of residues:          155
 # Start alignment!
 # Model    1   62.156
 # Model    2   57.567
 # Model    3   78.881
 # Model    4   76.093
 # Model    5   78.612
 # Model    6   90.198
 # Model    7   87.193
 # Model    8   73.144
 # Model    9   71.957
 # Model   10  100.461
 # Model   11  112.469
 # Model   12  108.363
 # Model   13  139.715
 # Model   14  130.736
 # Model   15  123.928
 # Model   16  116.492
 # Model   17   34.216
 # Model   18   35.854
 # Model   19   27.489
 # Model   20   44.658
 # Model   21   45.993
 # Model   22   52.989
 # Model   23   42.962
 # Model   24   34.723
 # Model   25  110.321
... 
 # Model  158   59.206
 # Model  159   40.211
 # Model  160   27.201
 # write models into one .pdb file

End the interactive session:

[user@cn4213 ~]$ exit
[user@biowulf ~]$