hesther / espsim Goto Github PK

Hi,

I really like the ESP-Sim package, but I would like to speed up the calculations for larger datasets. I see that ESP-Sim only uses one CPU core on my machine. Is there an easy way to set the number of cores I want to use for ESP-Sim calculations?

(Sorry if this is a noob question)

Thank you in advance!

Robin

Hey Hesther,

Looking at all the appreciation and its presence in various other sources, I feel at loss not being able to use it.
Can you please help me installing it in Google Colab?

Thanks in advance
-Hemant

Hi @hesther ,

Appeciate your work. I am having difficulty understaning the esp similarity formula in the end of short_demonstration.ipynb.

It seems to focus on the sign alignment situation among all space, then why not integral of phi_a * phi_b / (|phi_a|*|phi_b|), maybe I get it wrong. What's the intuition behind the formula.

Hello @hesther, thank you for this nice tool!
I would like to try ML-based charges, but it doesn't work for me:

752 if train_args.features_scaling != predict_args.features_scaling:
    753     raise ValueError('If scaling of the additional features was done during training, the '
    754                      'same must be done during prediction.')
    756 # If atom descriptors were used during training, they must be used when predicting and vice-versa

AttributeError: 'Namespace' object has no attribute 'features_scaling'

Am I missing some reqirements?
Thank you++

From the tutorial I have been using the following code block to view the alignment of my prbMol to my refMol.
p = py3Dmol.view(width=400,height=400) dt = {} for i in range(len(prbMols)): dt[i] = [prbMols[i], refMol] interact(draw, ms=dt,p=fixed(p),confIds=fixed([6,6]));

However, there appears to be no clear way to select the confIds that have given the best simShape or simEsp score, and instead you have to manually iterate through the confIds until you see a pair that give reasonable overlay. Is there anyway to automatically retrieve the confID of the 'best scoring' conformations?

Thanks,
Noah

Hi, thanks for this great package!

I wanted to speed it up for use in large-scale screening, so vectorized the gaussian integration step:

Here's a minimal-ish example, taken from the scripts dir:
setup:

## Setup:

from rdkit import Chem
from rdkit.Chem import AllChem
from rdkit.Chem import Draw
from rdkit.Chem.Draw import IPythonConsole

from espsim import EmbedAlignConstrainedScore,ConstrainedEmbedMultipleConfs,GetEspSim, helpers

import numpy as np
from scipy.spatial.distance import cdist


## set up molecules:

refSmiles=['C1=CC=C(C=C1)C(C(=O)O)O','CCC(C(=O)O)O','OC(C(O)=O)c1ccc(Cl)cc1','C1=CC(=CC=C1C(C(=O)O)O)O','COc1ccc(cc1)C(O)C(O)=O','OC(C(O)=O)c1ccc(cc1)[N+]([O-])=O','CCCC(C(=O)O)O','CCC(C)C(C(=O)O)O','CC(C(=O)O)O']
prbSmile='C(C(C(=O)O)O)O'
refMols=[Chem.AddHs(Chem.MolFromSmiles(x)) for x in refSmiles]
prbMol=Chem.AddHs(Chem.MolFromSmiles(prbSmile))

patt=Chem.MolFromSmiles("[H]OC([H])(C)C(=O)O[H]",sanitize=False)
helper=Chem.AddHs(Chem.MolFromSmiles("[H]OC([H])(C)C(=O)O[H]"))

AllChem.EmbedMolecule(helper,AllChem.ETKDG()) #Embed first reference molecule, create one conformer
AllChem.UFFOptimizeMolecule(helper) #Optimize the coordinates of the conformer
core = AllChem.DeleteSubstructs(AllChem.ReplaceSidechains(helper,patt),Chem.MolFromSmiles('*')) #Create core molecule with 3D coordinates
core.UpdatePropertyCache()

# align the molecules:
simShape,simEsp=EmbedAlignConstrainedScore(prbMol,refMols,core)

vectorized integration function (should just be a drop-in for GaussInt):

## Define vectorized gaussian integration functions:
def VecGI(dist, charge1,charge2,):
    
    #These are precomputed coefficients:
    a=np.array([[ 15.90600036,   3.9534831 ,  17.61453176],
                [  3.9534831 ,   5.21580206,   1.91045387],
                [ 17.61453176,   1.91045387, 238.75820253]])
    b=np.array([[-0.02495   , -0.04539319, -0.00247124],
                [-0.04539319, -0.2513    , -0.00258662],
                [-0.00247124, -0.00258662, -0.0013    ]])
    
    a_flat = a.flatten()
    b_flat = b.flatten()
    dist = (dist**2).flatten()
    charges = (charge1[:,None]*charge2).flatten()
    
    return ((a_flat[:,None] * np.exp(dist * b_flat[:,None])).sum(0) * charges).sum()


def vecSim(refCoor, prbCoor, refCharge, prbCharge, metric):
    distPrbPrb = cdist(prbCoor,prbCoor)
    distPrbRef = cdist(prbCoor,refCoor)
    distRefRef = cdist(refCoor,refCoor)
    
    intPrbPrb= VecGI(distPrbPrb,prbCharge,prbCharge)
    intPrbRef= VecGI(distPrbRef,prbCharge,refCharge)
    intRefRef= VecGI(distRefRef,refCharge,refCharge)
    return SimilarityMetric(intPrbPrb,intRefRef,intPrbRef,metric)

test equivalence:

prbCoor = prbMol.GetConformer(0).GetPositions()
prbCharge = np.array([a.GetDoubleProp('_GasteigerCharge') for a in prbMol.GetAtoms()])

simEsp_vectorized = []

for refMol in refMols:
    refCoor = refMol.GetConformer(0).GetPositions()
    refCharge = np.array([a.GetDoubleProp('_GasteigerCharge') for a in refMol.GetAtoms()])
    
    metric = 'tanimoto'
    tanimoto_similarity = GetEspSim(prbMol, refMol, metric=metric, partialCharges='gasteiger')
    vectorized_tanimoto_similarity = vecSim(refCoor, prbCoor, refCharge, prbCharge,metric)

    print(np.isclose(tanimoto_similarity, vectorized_tanimoto_similarity), tanimoto_similarity, vectorized_tanimoto_similarity)
    

output:

True 0.6460256327061207 0.6460256327061223
True 0.6882976768462027 0.6882976768461678
True 0.6464181000546018 0.6464181000546553
True 0.4060487492735462 0.4060487492735308
True 0.30699421499052204 0.3069942149905257
True 0.23382614291885506 0.23382614291884982
True 0.6689556377160998 0.668955637716057
True 0.7182699415095307 0.718269941509474
True 0.7087896849268721 0.7087896849268774

These take ~6ms vs 100µs. The VecGI should just be a drop in replacement for GaussInt. Seems to work for Carbo and Tanimoto, so if it's of interest Im happy to submit a PR.
cheers
Lewis

Hi, thank you for the great work.

I meed some problem when I try to run through the exaple notebook with partialCharges='resp'. The error is about key-values of options in helper.py. I try to modify it by removing the list in line 82. Then the charge output becomes only one value rather than a list/array in line 51 of electrostatics.py. I am not an expert of Psi4, hope these description can help you understand the issue.

Appreicate it if you can help, :D!

Best,

Chao

Hello, thanks for this great package!

Do you have any plans to make this package available through conda or pip?

Tharindu

hello esther,

figure 10 in your paper is really nice but I didn't see the code to draw that in this repo. Is that using some other software?

Best,
Min Htoo

Hi @hesther ,

Deepmind released a functional approximation, DM21, claiming they didn't break the fractional charge and spin constraints, which seems to essential to molecule DFT. But it is built with PYSCF, will you consider integrating it in this project?

Best,

Chao

Would that be possible to push a git tag or a pypi release?

I might use it to make a conda package of espsim.

ping @hesther

Thank you for an excellent, and easy to use, package. I was wondering if you could shed some light on the origin of these co-efficients, and their significance?

The installation fails to complete, even with a clean conda installation. Problems include resp and psi4 not being available for the modern version of python aka > 3.6

Hi-

I love the work you've done on this package. I've been toying around with it and noticed that it can provide stochastic results. Would you welcome a PR that would add seeds to all of the top level functions that gets passed to the conformer generation functions? This would be similar to the param in your function ConstrainedEmbedMultipleConfs but passed on in the rest of the functions as well.

Hi,

I was interested in using this library to compare docking poses generated using a consensus docking approach. Essentially, I would like to compare the electrostatic field similarity between different poses for a single compound, to use the similarity data to cluster the poses, in a similar fashion that you would do using RMSD values. I am just wondering if it is possible to calculate the field similarity without prior alignment of the two molecules, as the docking poses geometry would need to be conserved. Also the two molecules I would be comparing would be the same compound just in different conformations.

Thanks for any insight you may have,

Tony

Does espsim support this function:
3D shape and chemical similarity of two identical substructures from two molecule
Or any suggestions?

hello esther,

(I found a fix, please see my comment, but I don't know the root cause)
i've been trying out the espsim library on some molecules. the demo notebooks are very helpful and the analyses in the paper are also nicely done. great work!

i was trying to compare the ESP similarity of ATP against another known inhibitor of CDK2.
Specifically, I used the ATP co-crystal ligand from PDB ID 1B38,
(https://models.rcsb.org/v1/1b38/ligand?auth_seq_id=381&label_asym_id=C&encoding=sdf&filename=1b38_C_ATP.sdf)
and the Dinaciclib co-crystal ligand from PDB ID 5L2W
(https://models.rcsb.org/v1/5l2w/ligand?auth_seq_id=900&label_asym_id=C&encoding=sdf&filename=5l2w_C_1QK.sdf)

I am trying EmbedAlignScore() on these 2 mols, and the calculation works when I use gasteiger, mmff, ml, but not resp.
psi4 complains about this error:

RuntimeError: 
Fatal Error: RHF: RHF reference is only for singlets.
Error occurred in file: /build/source/psi4/src/psi4/libscf_solver/rhf.cc on line: 92
The most recent 5 function calls were:
psi::PsiException::PsiException(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, char const*, int)

I've also attached the detailed run log for your reference: https://gist.github.com/linminhtoo/84182da4bf727361b23905f34a429a5d

Here was how I run the code (I did rename some variables for my convenience but I didn't change any code logic)

# load co-crystal ATP
mol_atp = Chem.SDMolSupplier(str(RELATIVE / "data/cdk2_knownhits/1b38_C_ATP.sdf"), removeHs=False)[0]
mol_atp = Chem.AddHs(mol_atp, addCoords=True)

# load co-crystal dinaciclib
mol_dina = Chem.SDMolSupplier(str(RELATIVE / "data/cdk2_knownhits/5l2w_C_1QK.sdf"), removeHs=False)[0]
mol_dina = Chem.AddHs(mol_dina, addCoords=True)

# run ESPSim
shape_sims, esp_sims = EmbedAlignScore(
    probe_mol=deepcopy(mol_dina),
    ref_mols=deepcopy(mol_atp),
    probe_num_confs=10,
    ref_num_confs=10,
    partial_charge_mode="resp",
    renormalize=True,  # to [0, 1]
    getBestESP=True,  # more accurate but slower
    randomseed=2342,
)

I tried swapping probe_mol to mol_atp and ref_mols to mol_dina and psi4 did calculate the charges for one molecule before crashing again with the same error, which means it is not happy with Dinaciclib for some reason, but is fine with ATP. (assuming it calculates probe_mol first.

Do you happen to have any idea what's the issue?

Best,
Min Htoo

Hello,

I just downloaded this package and excited to use it. However, when I run benchmark_1_partial_charges.ipynb I'm getting this warning for the ML predictions:

Computing ML charges (will print three warnings for failed predictions)
Warning: could not obtain prediction, defaulting to Gasteiger charges for one molecule
Warning: could not obtain prediction, defaulting to Gasteiger charges for one molecule
Warning: could not obtain prediction, defaulting to Gasteiger charges for one molecule

Do I need to worry about this?

Thanks and have a nice day!

Trying to install ESPSim locally and conda is taking an age to solve the environment when using the .yml file. It is likely this is due to the .yml file not including specification of the Python or package versions required for a working install. Is it possible to upload a version which contains the dependencies and their versions?

Thanks,
Noah