overshiki / dlab-public

Code for the paper "DLAB - Deep learning methods for structure-basedvirtual screening of antibodies"

dlab-public

Code to run the antibody virtual screening pipeline described in the paper "DLAB - Deep learning methods for structure-based virtual screening of antibodies" (bioRxiv)[1]

Install

1. run bash build.sh

2. Get ZDock and copy the source code files into the folder external/zdock-3.0.2-src.

Test your install

The folder tests/ contains scripts you can use to check if your folder structure is set up correctly. Run them with pytest:

pytest tests/

Data preparation

1. Prepare your input antibody and antigen structures by running the mark_sur script as per the ZDock README. If you want to limit docking to the (predicted) interaction site, block atoms outside the interaction site by changing column 55-56 after running mark_sur to 19 as per the ZDock README. There is more detail on this in the paper.

2. Generate a .csv file containing all antibody and antigen pairings you want to investigate. An example is shown in example_input_files/pairings.csv.

3. Use the data_prep_pipeline.py script to run ZDock and generate types files. This uses a python implementation of the atomtyper functionality in libmolgrid [2]:

   python data_prep_pipeline.py -c data_prep_config.yaml
   

   data_prep_config.yaml configures the pipeline (see example_input_files/data_prep_config.yaml).

Running DLAB

For this script, you will need to be on a machine with GPU and CUDA, which is why this is seperate from the data preperation script (docking and preperation can be run on cpu compute servers before running GPU computations). Run DLAB (both rescoring and virtual screening in one go) using this command:

python dlab_re_vs_pipeline.py -c dlab_config.yaml

Where the yaml file defines the locations and types of models used as well as the input data and output file. An example can be found in example_input_files/dlab_config.yaml.

Analysing the output

The pipeline generates a csv file with the columns

• name
• dlab-re-max
• dlab-vs
• zdock_score

To follow the approach used in the paper ("DLAB-VS+ZDock"), a final score can be calculated by minmax scaling both DLAB-VS and ZDock score for each target antigen and averaging the two scores. For the approach denoted "DLAB-Re-max thresholding" in the paper, discard all antibody-antigen pairings not falling in the top 20% of DLAB-Re-max scores.

References

[1] DLAB - Deep learning methods for structure-based virtual screening of antibodies. C Schneider, A Buchanan, B Taddese, CM Deane, bioRxiv, 2021

[2] libmolgrid: Graphics Processing Unit Accelerated Molecular Gridding for Deep Learning Applications. J Sunseri, DR Koes. Journal of Chemical Information and Modeling, 2020