OPUS-Rota3

Side-chain modeling is critical for protein structure prediction since the uniqueness of the protein structure is largely determined by its side-chain packing conformation. In this paper, differing from most approaches which reply on rotamer library sampling, we first propose a novel side-chain rotamer prediction method based on deep neural networks, named OPUS-RotaNN. Then, on the basis of our previous work OPUS-Rota2, we propose an open-source side-chain modeling framework, OPUS-Rota3, which integrates the results of different methods into its rotamer library as the sampling candidates. By including OPUS-RotaNN into OPUS-Rota3, we conduct our experiments on three native backbone test sets and one non-native backbone test set. On the native backbone test set CAMEO-Hard61 for example, OPUS-Rota3 successfully predicts 51.14% of all side-chain dihedral angles with a tolerance criterion of 20°, outperforms OSCAR-star (50.87%), SCWRL4 (50.40%) and FASPR (49.85%). On the non-native backbone test set DB379-ITASSER, the accuracy of OPUS-Rota3 is 52.49%, better than OSCAR-star (48.95%), FASPR (48.69%) and SCWRL4 (48.29%).

Datasets

Train & Val:

The training and validation sets for OPUS-RotaNN are the same as that for OPUS-TASS, which are hosted on Baidu Drive with password zmc1. Also, they can be downloaded directly from Here. More details can be found in OPUS-TASS repo.

The CSF120 features for the training and validation sets are hosted on Baidu Drive with password 7bmi. Also, they can be downloaded directly from Here.

Test2016 & CASP-FM & CAMEO-Hard61

The pdb files of these three sets are hosted on Baidu Drive with password em4s. Also, they can be downloaded directly from Here.

DB379 & DB379-ITASSER:

These two sets can be downloaded from FASPR paper.

OPUS-RotaNN

Dependency

Python 3.7
TensorFlow v2.0
xssp-3.0.10

The training & inference codes of OPUS-RotaNN can be found Here.

The standalone version of OPUS-RotaNN (including pre-trained models, the tmp_files and the results for CASP-FM) is hosted on Baidu Drive with password 3l8k. Also, it can be downloaded directly from Here.

Note that for higher speed, we simplify the calculation in xssp. The pipeline for recompiling the xssp is as following:

cd mkdssp
tar zxvf xssp-3.0.10.tar.gz
cp dssp.cpp xssp-3.0.10/src/
cd xssp-3.0.10/
./autogen.sh
./configure
make mkdssp

OPUS-Rota3v and OPUS-Rota3

Dependency

For your convenience, the packages we used can be found in OPUS-Rota3/dependency folder.

redis
hiredis

Third-Party Programs

OSCAR-star

The executable oscar-star file we downloaded from OSCAR-star website can be found in OPUS-Rota3/oscar folder.

FASPR

The source code we downloaded from FASPR can be found in OPUS-Rota3/FASPR folder. For higher speed, we modified the source code. You can recompile it following their instruction.

OPUS-Rota2

The information of OPUS-Rota2 can be found here.

Usage

1. Database preparation.

   1. Install Redis.

      cd redis-3.2.12
      make
      cd src
      sudo make install
      

   2. Download OPUS-DASF lookup tables and rotamer library data, which are hosted on Baidu Drive with password 9dqu. Also, they can be downloaded directly from Here. Unzip dump.zip and put it into redis-3.2.12.

2. Install eigen3

   sudo apt install libeigen3-dev
   cd /usr/include/eigen3
   sudo cp Eigen/ .. -R
   

3. Install hiredis

   cd hiredis
   make
   sudo make install
   sudo ldconfig /usr/local/lib
   

4. Build OPUS-Rota3 (gcc/g++ 4.8 is necessary to reproduce our results)

   cd src
   mkdir build
   cd build
   cmake ..
   make
   cp rota3 ../../
   

5. Run Redis.

   cd /home/redis-3.2.12
   redis-server redis.conf
   

   This will take approximately 1min until "The server is now ready to accept connections on port 6379" appeared in the console.

6. Run OPUS-Rota3.

   Set the parameters in rota3.ini. The examples for OPUS-Rota3v and OPUS-Rota3 are rota3v_example.ini and rota3_example.ini, you can replace rota3.ini with what you need. Note that absolute path is required.

   1. optimized.

   optimized=0 means the sampling process will not be performed. If you want to use OPUS-Rota3v, please set optimized=0. For OPUS-Rota3, please set optimized=1.

   2. use_faspr

   If you want to use the results from FASPR, please set use_faspr=1. For both OPUS-Rota3v and OPUS-Rota3, use_faspr is 1.

   3. use_oscar

   If you want to use the results from OCAR-star, please set use_oscar=1.

   4. use_others

   Note that OCAR-star will provide stochastic results. Therefore, we provide another option that will not run OCAR-star on-the-fly, but introduce their existing predicted results. In this case, please set use_oscar=0 and use_others=1. The others_list.txt in tmp_dir should be maintained for establishing the relations between backbone files and OCAR-star predicted results. Instead of introducing the results from OCAR-star, the results from other side-chain modeling programs can also be included in this way. .dihedrals: real resid in backbone file, resname, chi1, chi2, chi3, chi4 (181: Missing atoms for calculation 182: chi doesn't exist).

   5. use_rotann

   If you want to use the results from OPUS-RotaNN, please set use_rotann=1 and maintain the rotann_list.txt in tmp_dir as use_others. Note that the different between use_others and use_rotann is .dihedrals uses real resid in backbone file and use_rotann uses the resid starting from 1. .rota: resid starting from 1, resname, chi1, chi2, chi3, chi4 (181: Missing atoms for calculation 182: chi doesn't exist).

Performance

Performance of different predictors on TEST2016

Predictors	MAE (χ1)	MAE (χ2)	MAE (χ3)	MAE (χ4)	ACC
OPUS-RotaNN	24.75	36.64	51.62	51.65	46.26%
RotamerLib	34.72	46.34	53.60	54.82	46.73%
SCWRL4	22.12	37.37	49.76	53.97	57.72%
FASPR	21.79	36.97	49.45	54.86	57.80%
OSCAR-star	19.74	35.85	47.90	53.30	58.99%

Performance of different predictors on DB379

Predictors	MAE (χ1)	MAE (χ2)	MAE (χ3)	MAE (χ4)	ACC
RotamerLib	34.47	45.38	56.50	53.28	46.82%
FASPR	21.52	36.41	53.09	53.45	57.98%
SCWRL4	21.99	36.89	52.84	52.55	58.03%
OSCAR-star	19.48	35.71	51.67	52.05	59.02%
OPUS-Rota3v	18.98	33.37	49.80	52.06	60.50%
OPUS-Rota3	17.52	32.85	48.96	50.42	61.52%

Performance of different predictors on CASP-FM

Predictors	MAE (χ1)	MAE (χ2)	MAE (χ3)	MAE (χ4)	ACC
OPUS-RotaNN	29.41	38.95	53.26	49.19	42.86%
RotamerLib	38.07	47.53	55.65	52.86	44.95%
FASPR	26.63	39.75	53.40	54.81	53.11%
SCWRL4	27.09	40.44	52.67	54.61	53.17%
OSCAR-star	24.63	37.68	50.61	53.35	54.84%
OPUS-Rota3v	23.05	36.05	50.82	53.63	56.65%
OPUS-Rota3	21.38	34.50	49.07	51.51	58.05%
OPUS-Rota3 (w/ RotaNN)	21.30	33.61	49.07	51.48	58.56%

Performance of different predictors on CAMEO-Hard61

Predictors	MAE (χ1)	MAE (χ2)	MAE (χ3)	MAE (χ4)	ACC
OPUS-RotaNN	32.73	42.18	56.37	51.01	38.51%
RotamerLib	41.39	50.25	58.52	54.44	40.44%
FASPR	28.47	42.03	55.52	56.91	49.85%
OPUS-Rota3v	28.15	41.97	55.76	57.07	50.24%
SCWRL4	28.26	42.35	55.96	55.97	50.40%
OSCAR-star	26.33	41.47	55.07	56.44	50.87%
OPUS-Rota3 (w/ RotaNN)	26.28	41.43	55.21	56.46	51.14%
OPUS-Rota3	26.29	41.48	55.26	56.49	51.16%

Performance of different predictors on DB379-ITASSER (Non-Native Backbone Test Set)

Predictors	MAE (χ1)	MAE (χ2)	MAE (χ3)	MAE (χ4)	ACC
RotamerLib	43.10	47.64	56.94	53.27	40.61%
SCWRL4	33.29	42.74	56.17	55.52	48.29%
FASPR	32.59	42.05	57.05	55.69	48.69%
OSCAR-star	30.78	42.45	56.07	55.74	48.95%
OPUS-Rota3v	28.33	38.84	53.64	54.49	51.88%
OPUS-Rota3	27.05	39.08	53.41	54.30	52.28%

Reference

@article{xu2020opus4,
  title={OPUS-Rota3: Improving Protein Side-Chain Modeling by Deep Neural Networks and Ensemble Methods},
  author={Xu, Gang and Wang, Qinghua and Ma, Jianpeng},
  journal={Journal of Chemical Information and Modeling},
  year={2020},
  publisher={ACS Publications}
}