ReactNetGen (Chemical Reaction Network Generator)

Enable ranking elementary reactions and generating reaction networks. The rank information was learned by an uncertainty calibrated learning to rank (LTR) model on the quantum chemical database and RMG-database.

• Uncertainty-calibrated deep learning approach for rapid identification of gas-phase reaction networks: peer-reviewed publication (open access).

Requirements

For basic use, installing:

• python
• numpy
• pandas
• pytorch
• rdkit
• sklearn
• tqdm
• networkx

RMG-database can be installed as the rank engine by installing

• RMG-Py (version == 3.1.0)
• RMG-database (version == 3.0.0)

Usage

Basic usage (Finding possible reaction pathways)

1. Import the NetGenerator and load reaction temperates for generating reaction pathways

from NetGenerator.rank_reactions import form_data, rank_reactions, show_results
import NetGenerator.GenNet as GenNet

react_dic, react_dic_reverse = GenNet.read_temp('./templates/rmg_template.txt')
reactant = GenNet.assign_map_num('C1CC[CH]C1=O')
outcomes = GenNet.run_one_rank(reactant, react_dic, react_dic_reverse, use_forward_temp=True, use_reverse_temp=True)

2. Format input file for ranking reaction pathways

reaction_df = form_data([reactant], outcomes)

3. Rank reaction pathways using LTR models

path = r'./reactranker/model_results/final_results/evidential_ranking' # load LTR models
sort_smile_with_map_and_scores, sort_smile = rank_reactions(path, reaction_df)

Basic usage (Generating reaction networks)

1. Importing the NetGenerator and load reaction temperates for generating reaction pathways

from NetGenerator.GenNet import NetGen, read_temp, trans_to_graph
dic_path = './templates/rmg_template.txt'
react_dic, react_dic_reverse = read_temp(dic_path)

2. Genertaing reaction networks

smi_list = ['C1C[CH]CC1=O', 'C1CC[CH]C1=O', '[CH2]CCC[C]=O']  # the reactant list
ranks = 4  # The species rank for terminating the generation of reaction networks
path = r'./reactranker/model_results/final_results/evidential_ranking'  # The LTR model path
cut_off = 2  # reserveing 2 reactions per rank
min_paths = 1  # reserveing 1 reactions at least per rank
use_uncertainty = True  # using uncertainty calibrated results
rank_pathways = True  # ranking the pathways
reaction_list, species_dic,info = NetGen(react_list=smi_list, ranks=ranks, cut_off=cut_off, min_paths=min_paths, 
                                    bi_molecule=False, react_dic=react_dic, react_dic_reverse=react_dic_reverse,
                                    use_uncertainty=use_uncertainty, use_reverse_temp=True, dropped_temp=[], 
                                    model_path=path, show_info=True, rank_pathways=True, using_rmg_database=False,
                                    rmg_estimator='rate rules')

Advanced usage

Template operation

Add new reaction template

To add new reaction template, following format should be added to templates\rmg_template.txt:

template_name: {"reaction_smarts":reaction_smart_string, 
                "react_temp_radical": {"atom map number":int}, 
                "prod_temp_radical": {"atom map number":int}}

Where template_name is the string format of the template name, reaction_smart_string is the string format of atom mapped reaction SMARTS.

Dropping reaction template for reaction pathway generation

The reaction template can be forbiden by specifying variable dropped_temp of NetGen and run_one_rank function

Adding LTR models

The user tranined LRT mdoel can be added to ReactNetGen\reactranker\model_results for ranking reaction pathways.

Ranking reaction pathways using RMG-database

RMG-database(version 3.0.0) can be installed as the rank engine. The reactions are ranked by RMG rate rules and RMG group additivity based on RMG-database(by specifying using_rmg_database).