Hi everyone,
I'm trying to reproduce the numbers I've found in this publication about GenScore .
I'm interested in the Enrichment Factor. I've started my analysis by downloading CASF-2016 from the official website.
The following is the process I followed to try and reproduce the data.
For each of the Binders of the CASF available inside CASF-2016/decoys_screening I've computed the scoring of all ligands inside each folder using the following command:
casf_output_ef.txt

python3.8 ./example/genscore.py -p CASF-2016/coreset/<target>/<target>_pocket.pdb -l CASF_2016/decoys_screening/<target>/ligands.mol2 -e gatedgcn -m ./trained_models/GatedGCN_0.5_1.pth -o <out_file>
Now that I have all the scores computed by GenScore, I've computed the enrichment factor with the following:
python2 CASF-2016/power_screening/forward_screening_power.py -c CoreSet.dat -s <genscore_results_folder> -t TargetInfo.dat -p 'positive' -o 'GenScore'
The problem I get is from the output of this script (I'm attaching the output I got):
Average enrichment factor among top 1% = 8.11 Average enrichment factor among top 5% = 3.36 Average enrichment factor among top 10% = 2.31 The best ligand is found among top 1% candidates for 15 cluster(s); success rate = 26.3% The best ligand is found among top 5% candidates for 24 cluster(s); success rate = 42.1% The best ligand is found among top 10% candidates for 28 cluster(s); success rate = 49.1%
Which is different from the number I've found in the publication

What am I missing?

Thank you in advance for all the help.

I have a hard time installing all the dependencies.

I am using cuda_11.5.2 toolkit in a docker container. However when installing the dependencies with pip or anaconda (python 3.8.11), I run into some compatibility problems between torch, torch-scatter...

Can someone help me with this, I feel like I am on the wrong cuda or python version?

Warm regards,
Wout

Hello, thank you for your incredible work!

I would like to do finetuning with some of your pre-trained models, what should I look at?
Could you give me an example of a command line string that would allow me to execute finetuning on the NN?

Best regards,
Vittorio.

Hi,
I ran GenScore on PDBbind-crossdock-core for testing (The model I tested was GT_ft_1.0_1.pth, using the default parameters for everything else.). However, for the decoys generated by vina, 505 receptor-ligand pairs (1343 in total) reported errors:
"""
Traceback (most recent call last):
File ".../GenScore/GenScore/example/genscore.py", line 256, in
main()
File ".../GenScore/GenScore/example/genscore.py", line 237, in main
ids, scores = scoring(prot=inargs.prot,
File ".../GenScore/GenScore/example/genscore.py", line 82, in scoring
data = VSDataset(ligs=lig,
File ".../GenScore/GenScore/example/../GenScore/data/data.py", line 178, in init
self.ids, self.gls = zip(*filter(lambda x: x[1] != None, zip(self.idsx, self.gls)))
ValueError: not enough values to unpack (expected 2, got 0)
"""
In all the cases where score can be generated, I calculated that the cross-docking SR1 is only 0.463 (0.59 reported in paper), and the SR1 of redocking is 0.680.

What problems do I need to correct when running genscore?

Best regards.

i run the dockies with example/genscore.py with target name '11betahsd1' have error:
gen_score_file(protein_path=protein_path,ligand_path=ligand_path,cand_sdf_path=cand_decoys_sdf,out_put_path=cand_decoys_score_outpath,args=args)
File "evalation_with_dockies.py", line 10, in gen_score_file
ids, scores = genscore.scoring(prot=protein_path,
File "/home/internal-GenScore/example/genscore.py", line 86, in scoring
data = VSDataset(ligs=lig,
File "/home/internal-GenScore/example/../GenScore/data/data.py", line 130, in init
self.gp = prot_to_graph(self.prot, cutoff)
File "/home/internal-GenScore/example/../GenScore/feats/mol2graph_rdmda_res.py", line 40, in prot_to_graph
res_coods = th.tensor(np.array([np.concatenate([res.atoms.positions, np.full((RES_MAX_NATOMS-len(res.atoms), 3), np.nan)],axis=0) for res in u.residues]))
File "/home/internal-GenScore/example/../GenScore/feats/mol2graph_rdmda_res.py", line 40, in
res_coods = th.tensor(np.array([np.concatenate([res.atoms.positions, np.full((RES_MAX_NATOMS-len(res.atoms), 3), np.nan)],axis=0) for res in u.residues]))
File "/opt/conda/lib/python3.8/site-packages/numpy/core/numeric.py", line 343, in full
a = empty(shape, dtype, order)
ValueError: negative dimensions are not allowed

cause the error is RES_MAX_NATOMS=24 smaller than res_max_natoms = max([len(res.atoms) for res in u.residues]).
what happened?

Hi,

I just have some queston when running GenScore:

• I noticed there is an option for explicit_H in the genscore.py file. Just wonder if setting explicit_H=True will change the program, since a lot of my ligands are charged molecules.
• Will --atom_contribution and --res_contribution make the final score more reliable?
• There are three files for GT_ft_0.5 in trained_models folder, which one should I use?

Thanks!

Hi team,

Thanks for the interesting program. I am having trouble installing the dependencies using the provided requirements_conda.txt.

This is the log:

conda install -c conda-forge --file requirements_conda.txt
Channels:
 - conda-forge
 - defaults
Platform: linux-64
Collecting package metadata (repodata.json): done
Solving environment: failed

PackagesNotFoundError: The following packages are not available from current channels:

  - kaleido==0.2.1=pypi_0
  - plotly==5.6.0=pypi_0
  - scripttest==1.3=pypi_0
  - tenacity==8.0.1=pypi_0

Current channels:

  - https://conda.anaconda.org/conda-forge
  - defaults

To search for alternate channels that may provide the conda package you're
looking for, navigate to

    https://anaconda.org

and use the search bar at the top of the page.

Do you know how to solve this problem?

Thanks!

Hi,

1. It took me some time to get a working environment. For sure, you should remove this from your read.me file:

conda create --prefix xxx --file ./requirements_conda.txt
pip install -r ./requirements_pip.txt

Here is how I was able to create the environment:

conda create -n genscore python=3.8.11
conda activate genscore
conda install nvidia/label/cuda-11.5.2::cuda-toolkit
conda install mdanalysis==2.0.0 -c conda-forge
conda install rdkit==2021.03.5 -c conda-forge
pip install prody==2.1.0
conda install numpy==1.20.3 -c conda-forge
pip install torch==1.11.0+cu115 torchvision==0.12.0+cu115 torchaudio==0.11.0 --extra-index-url https://download.pytorch.org/whl/cu115
pip install torch-geometric==2.0.3
pip install torch-scatter==2.0.9 -f https://pytorch-geometric.com/whl/torch-1.11.0+cu115.html
pip install seaborn==0.11.2
pip install pandas==1.3.2
pip install matplotlib==3.4.3
pip install scipy==1.6.2
pip install scikit-learn==0.24.2
conda install conda-forge::pytorch_sparse
pip install torch_sparse==0.6.15 -f https://pytorch-geometric.com/whl/torch-1.11.0+cu115.html
conda install openbabel==3.1.0 -c conda-forge
pip install joblib==1.0.1

2. in genscore.py, you hard-coded the link to openbabel libraries but you did not indicate it in your Read.me file

#you need to set the babel libdir first if you need to generate the pocket
#os.environ["BABEL_LIBDIR"] = "/home/shenchao/.conda/envs/my3/lib/openbabel/3.1.0"

You could replace this code by:

default_babel_libdir = os.path.join(os.getenv("CONDA_PREFIX", ""), "lib", "openbabel", "3.1.0")
os.environ["BABEL_LIBDIR"] = os.getenv("BABEL_LIBDIR", default_babel_libdir)

so it is not hard-coded anymore. However, it will still be linked to a variable... CONDA_PREFIX

3. I get deprecation warnings from MDAnalysis, Bio.pairwise2, and pkg_resources. Maybe other versions would solve the problem.

Best,
Christian

Can you please explain the output data? From the example data, I cannot find the output scores of the protein-ligand complex. Are higher scores better? Is there a way to convert to affinity?

Hi,
The pocket extraction step does not generate a pocket file. It seems to be doing something but no file is produced from it. If I put back the file: 1qkt_p_pocket_10.0.pdb that was already there before I ran the script: run_genscore.sh the scoring is done... so the only step that does not work is the pocket generation. One thing that does not seem to work is --parallel... I used "time python" to measure the execution time. It is exactly the same with or w/o the argument.

(genscore) christian@Linux00:/media/christian/VS1/VS/Tool_GenScore/example$ python genscore.py -p ./1qkt_p.pdb -l ./1qkt_decoys.sdf -rl ./1qkt_l.sdf -gen_pocket -c 10.0 -e gt -m ../trained_models/GT_0.0_1.pth
/home/christian/anaconda3/envs/genscore/lib/python3.8/site-packages/MDAnalysis/coordinates/chemfiles.py:108: DeprecationWarning: distutils Version classes are deprecated. Use packaging.version instead.
MIN_CHEMFILES_VERSION = LooseVersion("0.9")
/home/christian/anaconda3/envs/genscore/lib/python3.8/site-packages/MDAnalysis/analysis/data/filenames.py:82: DeprecationWarning: pkg_resources is deprecated as an API. See https://setuptools.pypa.io/en/latest/pkg_resources.html
from pkg_resources import resource_filename
/home/christian/anaconda3/envs/genscore/lib/python3.8/site-packages/Bio/pairwise2.py:278: BiopythonDeprecationWarning: Bio.pairwise2 has been deprecated, and we intend to remove it in a future release of Biopython. As an alternative, please consider using Bio.Align.PairwiseAligner as a replacement, and contact the Biopython developers if you still need the Bio.pairwise2 module.
warnings.warn(
@> 8026 atoms and 1 coordinate set(s) were parsed in 0.04s.
@> 44 atoms and 1 coordinate set(s) were parsed in 0.00s.

Hello! Could you offer the preprocessed datasets (lig.pt, prot.pt) mentioned in "train_model.py" or the download url of them? Thanks a lot.

run the model with GT_0.0_1.pth.
parameter:
args["batch_size"] = 128
rgs["dist_threhold"] = 5.0
args['device'] = 'cuda:6' if th.cuda.is_available() else 'cpu'
print(f"""{args['device']}""")
args["num_workers"] = 10
args["num_node_featsp"] = 41
args["num_node_featsl"] = 41
args["num_edge_featsp"] = 5
args["num_edge_featsl"] = 10
args["hidden_dim0"] = 128
args["hidden_dim"] = 128
args["n_gaussians"] = 10
args["dropout_rate"] = 0.15

genscore.scoring paramter cut=5.0

dekios2.0x test data:
protein and ref-ligand dir: aurka_prot

dekios2.0x decoys and actives dir:
aurka_decoys_SP
aurka_actives_SP

average auc is:0.593.
can you help me what's parameter Caused difference

Hi,
I am trying to figure out how to use GenScore based on your example: 1qkt. Could you give me some details about the files, please?
I guess you prepared the files (1qkt_p.pdb & 1qkt_l.sdf) from 1qkt so the estrogen nuclear receptor with its ligand estrogen, using Maestro. Did you get 1qkt from RCSB pdb or from pdb-redo? Did you apply some minimization on his complex during preparation?

How did you generate: 1qkt_decoys.sdf? Why is it used with -gen_pocket... the binding pocket is selected based on both 1qkt_l.sdf and 1qkt_decoys.sdf?

GenScore accepts ligand files only in sdf or also in other format such as mol2?

Thanks for your help