FP-Zernike: An open-source structural database construction toolkit for fast and accurate structure retrieval

The main functions of FP-Zernike as follows:

1. Given a protein(or RNA) structure(format: *.pdb), FP-Zernike can calculate its Zernike descriptors[4 types, namely PM-Zernike, PS-Zernike, ATOM-Zernike and GM-Zernike]. Based on the results of the experiment, we recommend the user to calculate PM-Zernike, PS-Zernike or ATOM-Zernike.

2. Given a customized dataset[only include *.pdb file], FP-Zernike can calculate all Zernike descriptors about it.

3. Given a query structure[single-chain], structure retrieval is done on a dataset containing 590685 protein chains[run time: ~4s-9s].

4. Based on the Euclidean distance of descriptors, the similarity of structures is preliminarily evaluated. When the Euclidean distance between descriptors is less than 2, it means that there is strong similarity.

5. Based on 'Romokage', the similarity between structures is measured.

Packages && deploy
In order to run FP-Zernike, three main packages are required, that is, 'pymol', 'numpy' and 'numba'. For convenience, an environment file('FP_Zernike.yaml') is provided so that users can deploy FP-Zernike directly using conda, the command line is as follows:
conda env create -f FP_Zernike.yaml
pip install scikit-learn
In addtion, users need to create a folder by command line as follows:
mkdir tempoutput

Usage
1. Calculate Zernike descriptors
Given the query structure[path: example/4mu3A.pdb], the user can execute the following command to get its ATOM-Zernike:
python mainZernikeCalculation.py -i example/4mu3A.pdb -file_source atom -mode atom -o test_4mu3A_ATOM.pkl
Here, 'test_4m2qA_ATOM.pkl' include the information of Zernike descriptor of 'example/4m2qA.pdb'.
The user can execute the following command to get its PM-Zernike:
python mainZernikeCalculation.py -i example/4mu3A.pdb -file_source pymol -mode mesh -o test_4mu3A_PM.pkl
The user can execute the following command to get its PS-Zernike:
python mainZernikeCalculation.py -i example/4mu3A.pdb -file_source pymol -mode surface -o test_4mu3A_PS.pkl
The user can execute the following command to get its GM-Zernike:
python mainZernikeCalculation.py -i example/4mu3A.pdb -file_source gmconvert -mode mesh -o test_4mu3A_GM.pkl
In addition, users can enter the following command to obtain detailed instructions.
python mainZernikeCalculation.py -h
The above commands in 'test_mainZernikeCalculation.sh', user can run 'bash test_mainZernikeCalculation.sh' to test this function.

2. Build a database of descriptors
Given the customized dataset[only include *.pdb file, path: example], the user can execute the following command to get its ATOM-Zernike database:
python mainMakeZernikeDataBase.py -i example -t 4 -mode ATOM
Here, '4' is the thread number. A new folder[path: exampleAtomZernikeDescriptor] containing all descriptors is automatically generated. Similarly, descriptors for other patterns can also be built, and these command lines are in 'test_mainMakeZernikeDataBase.sh', user can run 'bash test_mainMakeZernikeDataBase.sh' to test this function.
In addition, users can enter the following command to obtain detailed instructions.
python mainMakeZernikeDataBase.py -h

3. Ultra-fast structure retrieval

Given the query structure, the user can perform ultra-fast structure retrieval. For example, given a protein chain[Path:example/4mu1A.pdb], users can preform sturcture retrieval by command as follows.
python mainRetrieve.py -q example/4mu1A.pdb -dm PM -otn 10 -op testPMRtriRes.txt
Here, 'PM' is the mode of retrieve, '10' is the number of output structures, 'testPMRtriRes.txt' stores all the output.
In addition, users can enter the following command to obtain detailed instructions.
python mainRetrieve.py -h

There are more commands in 'test_mainRetrieve.sh', and users can test this function by executing 'bash test_mainRetrieve.sh'.

We constructed a demo containing 590685 structures, at this scale, our retrieval system only takes 4 - 9 seconds to complete a retrieval. To perform the retrieval on the demo dataset, users need to through the following link to download about demo database, and need to put it copy to the folder 'database'.

link: https://pan.baidu.com/s/1NnQsFiVyatQ2p-k7oj-giA
Extraction code: 208z

4. Measure the similarity between the two structures
When the descriptors of the structure are calculated by FP-Zernike, the user can initially measure the similarity based on the Euclidian distance between the descriptors. Specifically, given the descriptors of two structures, A[Path: example/4mr5A.pkl] and B[Path: example/4mr6A.pkl], the user can perform the following command to obtain the Euclidean distance between them.
python mainZernikeDist.py example/4mr5A.pkl example/4mr6A.pkl

5. Measure the similarity based on ReOmokage
Here, we improve the Omokage and propose 'ReOmokage' to measure the similarity between structures. Given structures A[Path: example/4mu3A.pdb] and B[Path: example/4mu1A.pdb], the user can enter the following command to calculate the similarity between the structures:
python mainReOmokage.py -pdb1 example/4mu3A.pdb -pdb2 example/4mu1A.pdb