Available Modules
Interactive Path Planning and Tree Builder
Interactive tool for manually and automatically exploring retrosynthesis predictions
This module allows you to build a retrosynthetic tree by combining multiple one-step retrosynthesis predictions. The interface also offers the ability to start automatic tree generation jobs using MCTS tree search algorithms combined with the same one-step retrosynthetic predictions.
The current one-step retrosynthesis model is template-based, using a neural network to predict the most relevant retrosynthetic templates for a given target molecule. (Chem. Eur. J. 2017, 23, 25, 5966-5971, J. Chem. Inf. Model. 2020, 60, 7, 3398–3407)
Retrosynthesis Prediction
Tool for exploring one-step retrosynthesis predictions from multiple models.
This module provides a convenient utility for running multiple one-step retrosynthesis predictions for a single target molecule. Results are displayed in a table for easily viewing common and unique suggestions from each model. This tool currently supports all of the template-based models available in the interactive path planner as well as additional model architectures via OpenRetro.
Reaction Condition Recommender
Predict reaction conditions for a given reaction
This module allows you to predict reagents, catalysts, solvents and temperature for a desired transformation using a neural network model. (ACS Cent. Sci., 2018, 4, 1465-1476)
The 2021.01 release adds a new version of the model which can provide quantitative condition predictions.
Forward Synthesis Predictor
Evaluate a reaction in the forward direction, considering reaction conditions and impurities
This module allows you to predict most likely outcomes of a chemical reaction using a template-free WLN model trained to predict likely bond changes. (Chem. Sci., 2019, 10, 370-377)
Reaction Impurity Prediction
Predict likely reaction impurities
This module allows you to predict likely impurities for a chemical reaction. The algorithm considers minor products, over-reaction, dimerization, solvent adducts, and subsets of reactants. It relies on the template-free forward prediction model. (Chem. Sci., 2019, 10, 370-377)
Universal Regioselectivity Prediction
Predict selectivity of regio-selective reactions
This module allows you to predict selectivity of regio-selective reactions. The QM-GNN model combines a WLN graph encoding with predicted quantum descriptors as input to a multitask neural network. (Chem. Sci., 2021, 12, 2198-2208)
Aromatic C-H Site Selectivity Prediction
Predict reactive aromatic C-H sites
This module allows you to predict site selectivity of 123 aromatic C-H functionalization reactions with a multitask neural network that uses a WLN graph encoding. (React. Chem. Eng., 2020, 5, 896-902)
SCScore Evaluator
Analyze a chemical's perceived synthetic complexity
This module is a utility to evaluate the SCScore (a learned synthetic complexity metric) for arbitrary target chemicals. The model was trained on reactions in Reaxys in an attempt to understand the nuance of synthetic complexity, e.g., when protections and deprotections are constructive. It is designed to perceive a monotonic increase in complexity throughout a linear synthesis. (J. Chem. Inf. Model. 2018, 58, 2, 252-261)
Search for a molecule SMILES on the homepage to evaluate its SCScore.
Reaction Classification
Label reaction types using categorization from NameRxn
This module assigns human-interpretable reaction types for arbitrary reactions. It is based on the BERT natural language processing model, and is trained on labeled reactions from NameRxn in Pistachio. (ChemRxiv 2019)
Search for a reaction SMILES on the homepage to classify it, or use the reaction class cluster method in the Interactive Path Planner.
Solubility Prediction
Predict solid solubility and related properties
This module allows you to predict solid solubility for solvent/solute pairs at a desired temperature. Additional properties are also returned, including free energy and enthalpy of solvation, Abraham solute parameters, and others. The model relies on multiple machine learning models as well as thermodynamic relationships to derive the final result. Prediction uncertainties are also estimated by using a model ensemble approach. (ChemRxiv 2022)
Solvent Screening
Compare solubility across sets of process solvents
This module leverages the solubility prediction model to provide a convenient tool for screening sets of process solvents across multiple temperatures. Results can currently be visualized using a bar chart to compare across different solvents or a line chart to visualize temperature dependence. The generated charts and data can be easily saved in various formats. (ChemRxiv 2022)
Buyable Look-up
Check to see if a chemical is in the buyables database
This module is a utility to check if a certain compound is included in the building block database. By default, the database includes approximately 280,000 database entries from eMolecules, LabNetwork, and Sigma Aldrich with prices up to $100 per gram.
Users can also add custom building blocks to the database, which will be reflected in the Interactive Path Planner and Tree Builder modules.
Atom Mapping
Generate atom mapping for a reaction SMILES
This module provides a utility for generating atom mappings for reactions. Currently, there are two models available: a WLN atom mapper using the template-free forward prediction model, (Chem. Sci., 2019, 10, 370-377) and the RXNMapper transformer model. (ChemRxiv 2020)
Search for a reaction SMILES on the homepage to generate an atom mapping.
Drawing
Draw molecules, reactions, or templates from SMILES/SMARTS strings
This module is a utility to draw SMILES strings describing chemicals, reactions, or SMARTS strings describing templates. Note that you can also copy a SMILES string and "Paste special" in Chemdraw (highly recommended), although this does not work very well for templates.