bruceunx/retroapi

RetroSynthesis API

• retroapi get predict routes given target molecule.

• supply plausible value for all predicts.

• predict conditions base on one reaction.

• add aync support

• deploy your backend server with torchserve (if api failed)

  install torchserve and other libraries from requirements.txt run download_models.sh run

  torchserve --start --foreground --ncs --model-store=mars --models reaxys=reaxys.mar
  

  test server

   curl http://127.0.0.1:8080/predictions/reaxys \                                                                                                            (base)
                --header "Content-Type: application/json" \
                --request POST \
                --data '{"smiles": ["CC(C)(C)OC(=O)N1CCC(OCCO)CC1"]}'

Install

• using pip

pip install retroapi

Usage

from retroapi import RetroApi, Name2Smiles

token = "" # you may get token first

retro_api = RetroApi(token)

# if you have simles of molecule.
smiles = "COc1cccc(OC(=O)/C=C/c2cc(OC)c(OC)c(OC)c2)c1"

# else:
chemical_name = "4-Hydroxycoumarin"
name2smiles = Name2Smiles()
smiles = name2smiles.get_smiles(chemical_name)

# check if smiles is valid or not
is_valid_smiles = retro_api.validate_smiles(smiles)

if is_valid_smiles:
    routes = retro_api.predict_routes(smiles)
    if routes is not None:
        # work with routes
        pass

# check if chemical is buyable or not
is_buyable = retro_api.check_stock(smiles)

# check reaction conditions

# first you should get product smile
# second you should get reactants smile

products = "COc1cc(C(=O)O)cc(OC)c1OC"
reactants = "C=CC(=O)O.COc1cc(Br)cc(OC)c1OC"

conds = retro_api.process_reaction(product, reactants)
if conds is not None:
    # check reaction condition with plausible
    pass

Async Usage

from retroapi import RetroApi, Name2Smiles


async def foo():
    retro_api = RetroApi(token)

    # if you have simles of molecule.
    smiles = "COc1cccc(OC(=O)/C=C/c2cc(OC)c(OC)c(OC)c2)c1"

    # else:
    chemical_name = "4-Hydroxycoumarin"
    name2smiles = Name2Smiles()
    smiles = await name2smiles.aget_smiles(chemical_name)

    # check if smiles is valid or not
    is_valid_smiles = await retro_api.avalidate_smiles(smiles)

    if is_valid_smiles:
        routes = await retro_api.apredict_routes(smiles)
        if routes is not None:
            # work with routes
            pass

    # check if chemical is buyable or not
    is_buyable = await retro_api.acheck_stock(smiles)

    # check reaction conditions

    # first you should get product smile
    # second you should get reactants smile

    products = "COc1cc(C(=O)O)cc(OC)c1OC"
    reactants = "C=CC(=O)O.COc1cc(Br)cc(OC)c1OC"

    conds = await retro_api.aprocess_reaction(product, reactants)
    if conds is not None:
        # check reaction condition with plausible
        pass

Change log:

2023-12-30 Extract condition recommend code from ASKCOS to notebook for local test.

2023-12-25 Run backend server with torchserve

Add try_times in predict_routes and process_reaction w/ async

routes = await retro_api.apredict_routes(smiles, try_num=20)

Need Token to use this package

It's wrap package for askcos.mit.edu API, so you can get token from website first.

add aync for package

with function name prefix with 'a', for example get_smile -> aget_smile

API Documentation

Class Name2Smiles

• get_smiles
  • parameter
    • chemical_name: str
  • output
    • smiles: str

Class RetroApi

• validate_smiles

  • parameter
    • smiles: str
  • output
    • true/false

• predict_routes

  • parameter
    • smiles: str
  • output
    • routes: list

• process_reaction

  • parameter
    • product: str

      product smiles

    • reactants: str

      reactants smiles jointed by "." like "C=CC(=O)O.COc1cc(Br)cc(OC)c1OC"

  • output conditions: list