Want to add custom scoring functions

[Shiv2116]

hey @halx can you guide me up on how to use the externalprocess component as i am planning to incorporate
''BA, SA, QED, Diversity, similarity, Molecular weight, MPO score, num rotatable bonds, , HBD, HBA, TPSA,
LogP, PLogP''
how would i incorporate this ? what shall i touch to do this? and how can i do this

[halx]

You may not need ExternalProcess as we cover (most) of this. What is BA and how do you plan to use similarity/diversity? We have a diversity filter available. QED covers the last five or so you mention btw. MPO usually stands for "multi parameter optimisation" so is a rather generic term.

As for ExternalProcess, there are several examples in support/.

[Shiv2116]

Okay so we are looking to use reinvent in our project, we wanna use our custom scores to be integrated
like through api call where will be calling reinvent to score and accordingly make our reward fucntions

i was looking for a way to intergrate the api into reinvent (figuring out still- help me on this)

like we will use the above scores that will hit reinvent and score us
Ba is Binding affinity

can you help me out in this?
where to intergrate , how to do this?

[Shiv2116]

Hey @halx awaiting your reply!!

also I needed to ask
how can we integrate custom prior model?

[halx]

For predictive models we have interfaces to ChemProp 1.x and QSARtuna. For a WebAPI we have comp_generic_rest.py but really it is just a matter of providing SMILES in a suitable format to the backend service and the scoring component then returns a score for each of the SMILES.

What exactly do you mean by "custom prior model"? If you are reusing the network architecture of the current code then it should be only a matter of reading in the model file, assuming you have followed the workflow in reinvent/runmodes/create_model. If you want to change the network architecture you willl need to write your own code.

[Shiv2116]

For the custom scoring part

we are defining our own molgen script
wanted to know how can we integrate that with the reward function reinvent uses
so that it can be replaced/ integrated through api call and hits reinvent whenever we wanna get molgen scores
?

[Shiv2116]

awaiting your response @halx !!

[halx]

I am happy to help but you need to remember that you are receiving here free user support for free software from a commercial entity. Thank you for your understanding.

[halx]

I think you need to be clearer on what you want to do. I do not know what that "molgen script" is and how that relates to scoring and why API calls are needed?

[Shiv2116]

okay let's keep everything aside
just taking an example

when we run reinvent_demo
it tries to optimise QED score

how can I make a file by passing the priors
how can I calculate binding affinity for it?

how can it optimise BA scores in every cycle?

[halx]

Regards, adding predictive models I have answered above. I do not understand why you need to pass the prior anywhere else and why that would need a file.

[Shiv2116]

See for eg i have written a custom tanimoto

from __future__ import annotations

__all__ = ["Tanimoto"]

from typing import List
import numpy as np
from rdkit import Chem
from rdkit.Chem.Scaffolds import MurckoScaffold
from rdkit.Chem.rdFingerprintGenerator import GetMorganGenerator
from rdkit.DataStructs import BulkTanimotoSimilarity
from .component_results import ComponentResults
from reinvent_plugins.mol_cache import molcache
from .add_tag import add_tag

@add_tag("__component")
class Tanimoto:
    def __init__(self,*args,**kwargs):
        self.morgan_generator = GetMorganGenerator(radius=2, fpSize=2048)

        self.prior_art = {
            '8GOD': {'smiles': 'Cc1c(nc2n1ccc(c2)C(=O)N3CCCC(C3)N)c4cc5ccccc5n4Cc6ccc(cc6)F'}
        }
        
        for key in self.prior_art.keys():
            self.prior_art[key]['mol'] = Chem.MolFromSmiles(self.prior_art[key]['smiles'])
            self.prior_art[key]['generic_scaffold'] = MurckoScaffold.MakeScaffoldGeneric(self.prior_art[key]['mol'])
            self.prior_art[key]['fp'] = self.morgan_generator.GetFingerprint(self.prior_art[key]['mol'])
            self.prior_art[key]['scaffold_fp'] = self.morgan_generator.GetFingerprint(self.prior_art[key]['generic_scaffold'])

    @molcache
    def __call__(self, mols: List[Chem.Mol]) -> np.ndarray:
        mol_generic_fps = []
        invalid_mols = []    
        for ind, m in enumerate(mols):
            try:
                mol_scaffold = MurckoScaffold.MakeScaffoldGeneric(m)
                mol_generic_fps.append(self.morgan_generator.GetFingerprint(mol_scaffold))
            except Exception as e:
                print(f"Error processing molecule {ind}: {e}")
                invalid_mols.append(ind)

        scores = list(BulkTanimotoSimilarity(self.prior_art['8GOD']['scaffold_fp'], mol_generic_fps))

        for idx in sorted(invalid_mols):  # Sort to maintain order
            scores.insert(idx, 0)
        
        return  ComponentResults(scores)                

for this I am facing this issue

Traceback (most recent call last):

  File "/opt/conda/envs/reinvent4/bin/reinvent", line 8, in <module>

    sys.exit(main_script())

  File "/opt/conda/envs/reinvent4/lib/python3.10/site-packages/reinvent/Reinvent.py", line 195, in main_script

    main(args)

  File "/opt/conda/envs/reinvent4/lib/python3.10/site-packages/reinvent/Reinvent.py", line 164, in main

    runner(

  File "/opt/conda/envs/reinvent4/lib/python3.10/site-packages/reinvent/runmodes/RL/run_staged_learning.py", line 363, in run_staged_learning

    terminate = optimize(package.terminator)

  File "/opt/conda/envs/reinvent4/lib/python3.10/site-packages/reinvent/runmodes/RL/learning.py", line 121, in optimize

    results = self.score()

  File "/opt/conda/envs/reinvent4/lib/python3.10/site-packages/reinvent/runmodes/RL/learning.py", line 195, in score

    results = self.scoring_function(

  File "/opt/conda/envs/reinvent4/lib/python3.10/site-packages/reinvent/scoring/scorer.py", line 143, in compute_results

    transform_result = compute_transform(

  File "/opt/conda/envs/reinvent4/lib/python3.10/site-packages/reinvent/scoring/compute_scores.py", line 101, in compute_transform

    component_results = compute_component_scores(

  File "/opt/conda/envs/reinvent4/lib/python3.10/site-packages/reinvent/scoring/compute_scores.py", line 67, in compute_component_scores

    for data in zip(smilies_non_cached, *component_results.scores):

TypeError: 'float' object is not iterable

[Shiv2116]

Hey @halx figured out the above thing

now i am looking into adding a custom prior
like this-
https://github.com/Rose-STL-Lab/LIMO
can you give advice how can I do this and pass this model integrated into reinvent

[Shiv2116]

@halx waiting for your response?
wanted to know if above thing can be done?

[halx]

A prior is a pretrained model. What you want is to integrate a whole new network architecture. You will need to look into reinvent/models in particular model_factory/model_adapter.py and see if you can shoehorn that code into it.

[Shiv2116]

i am facing a new error while making a new scoring function for this toml

stages = """


[[stage]]


max_score = 1.0

max_steps = 400


chkpt_file = 'stage1.chkpt'


[stage.scoring]

type = "geometric_mean"


[[stage.scoring.component]]

[stage.scoring.component.custom_alerts]


[[stage.scoring.component.custom_alerts.endpoint]]

name = "Alerts"


params.smarts = [

    "[*;r8]",

    "[*;r9]",

    "[*;r10]",

    "[*;r11]",

    "[*;r12]",

    "[*;r13]",

    "[*;r14]",

    "[*;r15]",

    "[*;r16]",

    "[*;r17]",

    "[#8][#8]",

    "[#6;+]",

    "[#16][#16]",

    "[#7;!n][S;!$(S(=O)=O)]",

    "[#7;!n][#7;!n]",

    "C#C",

    "C(=[O,S])[O,S]",

    "[#7;!n][C;!$(C(=[O,N])[N,O])][#16;!s]",

    "[#7;!n][C;!$(C(=[O,N])[N,O])][#7;!n]",

    "[#7;!n][C;!$(C(=[O,N])[N,O])][#8;!o]",

    "[#8;!o][C;!$(C(=[O,N])[N,O])][#16;!s]",

    "[#8;!o][C;!$(C(=[O,N])[N,O])][#8;!o]",

    "[#16;!s][C;!$(C(=[O,N])[N,O])][#16;!s]"

]


[[stage.scoring.component]]

[stage.scoring.component.QED]

[[stage.scoring.component.QED.endpoint]]

name = "QED"

weight = 1


[[stage.scoring.component]]

[stage.scoring.component.SAScore]

[[stage.scoring.component.SAScore.endpoint]]

name = "SA score"

weight = -0.5


[[stage.scoring.component]]

[stage.scoring.component.Tanimoto]

[[stage.scoring.component.Tanimoto.endpoint]]

name = "Tanimoto similarity"

weight = 3


[[stage.scoring.component]]

[stage.scoring.component.MolScore]

[[stage.scoring.component.MolScoreComponent.endpoint]]

Name = "MolScore"

for

from __future__ import annotations


__all__ = ["MolScoreComponent"]


import os

import json

import tempfile

import numpy as np

from typing import List

from .component_results import ComponentResults

from ..normalize import normalize_smiles

from .add_tag import add_tag

from molscore import MolScore  



@add_tag("__component")

class MolScore:

    def __init__(self, model_name: str = "scoring_model"):

        self.model_name = model_name

        self.scoring_config ={

  "task": "custom_score_qed_sa_tanim",

  "output_dir": "./",

  "load_from_previous": false,

  "logging": false,

  "monitor_app": false,

  "termination_exit": false,

  "scoring_functions": [

    {

      "name": "MolecularSimilarity",

      "run": true,

      "parameters": {

        "prefix": "8GOD",

        "ref_smiles": [

        "Cc1c(nc2n1ccc(c2)C(=O)N3CCCC(C3)N)c4cc5ccccc5n4Cc6ccc(cc6)F"

        ],

        "fp": "ECFP4",

        "bits": 1024,

        "similarity_measure": "Tanimoto",

        "method": "mean",

        "n_jobs": 1,

        "timeout": 60

      }

    },

    {

      "name": "MolecularDescriptors",

      "run": true,

      "parameters": {

        "prefix": "desc",

        "n_jobs": 1

      }

    }

  ],

  "scoring": {

    "metrics": [

      {

        "name": "8GOD_Sim",

        "filter": false,

        "weight": 1.0,

        "modifier": "norm",

        "parameters": {

          "objective": "maximize",

          "max": 1.0,

          "min": 0.0

        }

      },

      {

        "name": "desc_QED",

        "filter": false,

        "weight": 1.0,

        "modifier": "norm",

        "parameters": {

          "objective": "maximize",

          "max": 1.0,

          "min": 0.0

        }

      },

      {

        "name": "desc_SAscore",

        "filter": false,

        "weight": 1.0,

        "modifier": "norm",

        "parameters": {

          "objective": "minimize",

          "max": 1.0,

          "min": 0.0

        }

      }

    ],

    "method": "amean"

  },

  "diversity_filter": {

    "run": false

  }

        }


    @normalize_smiles

    def __call__(self, smiles: List[str]) -> ComponentResults:

        with tempfile.NamedTemporaryFile(mode='w+', suffix='.json', delete=False) as tmp_file:

            json.dump(self.scoring_config, tmp_file, indent=4)

            tmp_file_path = tmp_file.name


        try:

            ms = MolScore(model_name=self.model_name, task_config=tmp_file_path)

            scores = ms.score(smiles)

            json_data = json.loads(ms.main_df.to_json(orient='index'))


            all_results = {

                "scoring_results": {key: value for key, value in json_data.items()},

                "scores": scores.tolist() if isinstance(scores, (np.ndarray, np.generic)) else scores

            }

        finally:

            os.remove(tmp_file_path)  # Ensure cleanup


        return ComponentResults([all_results["scoring_results"], all_results["scores"]])

facing this error

File "/opt/conda/envs/reinvent4/bin/reinvent", line 8, in <module>
    sys.exit(main_script())
  File "/opt/conda/envs/reinvent4/lib/python3.10/site-packages/reinvent/Reinvent.py", line 195, in main_script
    main(args)
  File "/opt/conda/envs/reinvent4/lib/python3.10/site-packages/reinvent/Reinvent.py", line 164, in main
    runner(
  File "/opt/conda/envs/reinvent4/lib/python3.10/site-packages/reinvent/runmodes/RL/run_staged_learning.py", line 296, in run_staged_learning
    packages = create_packages(reward_strategy, stages, rdkit_smiles_flags2)
  File "/opt/conda/envs/reinvent4/lib/python3.10/site-packages/reinvent/runmodes/RL/run_staged_learning.py", line 157, in create_packages
    scoring_function = Scorer(stage.scoring)
  File "/opt/conda/envs/reinvent4/lib/python3.10/site-packages/reinvent/scoring/scorer.py", line 80, in __init__
    self.components = get_components(config.component)
  File "/opt/conda/envs/reinvent4/lib/python3.10/site-packages/reinvent/scoring/config.py", line 47, in get_components
    endpoints: dict = component_value["endpoint"]
KeyError: 'endpoint'

i didn't face this issue while introducing while custom tanimoto but facing this endpoint error when i introduced molscore

[halx]

I think you need to figure out yourself what is going wrong here. I cannot provide support for this. It also looks like a rather bad idea what you are trying to do there.