Introduction
Molecular properties play a crucial role in the design of materials for a wide range of energy applications. For instance, designing battery electrolytes requires satisfying key molecular characteristics: the temperature range of the liquid phase (i.e. melting and boiling points), oxidative stability (highest occupied molecular orbital) and reductive stability (lowest unoccupied molecular orbital), solubility (dipole moment), and flammability (flash point).
Model Overview
The Aionics models take as input the SMILES string of any molecule and employ message-passing (graph convolution) layers to extract graph-level features, which are then used to predict molecular properties. GNNs learn representations directly from molecular graphs to enable better predictive power than conventional feature-based models. Each prediction is accompanied by an estimate of the standard deviation, providing a measure of the model's uncertainty in making that prediction.
Dataset
All models are trained on proprietary Aionics datasets containing a mix of internal data and publicly available datasets from experiments and quantum chemistry calculations.
Models
These proprietary models are built by Aionics utilizing methodology developed in Prof. Venkat Viswanathan’s lab at Carnegie Mellon University as part of the ARPA-E DIFFERENTIATE program . The intellectual property from CMU is exclusively licensed to Aionics and represents one of the most advanced molecular property optimization capabilities for electrolyte design.
The six property models currently available for use are:
Melting Point
The Melting Point (MP) indicates the lowest temperature at which the electrolyte solvent remains in a liquid state. The MP model is trained on ~104 experimental data points and predicts the temperature of melting in degrees Celsius.
Boiling Point
The Boiling Point (BP) indicates the highest temperature at which the electrolyte solvent remains in a liquid state. The BP model is trained on ~104 experimental data points and predicts the temperature of boiling in degrees Celsius.
Flash Point
The Flash Point (FP) is a measure of the flammability of a solvent or additive. The FP model is trained on ~104 experimental data points and predicts the lowest temperature, in degrees Celsius, at which a solid or liquid can produce enough vapor to form a flammable air-vapor mixture.
Dipole Moment
The Dipole Moment provides a useful measure of the solubility of an additive molecule in the electrolyte solvent. The Dipole Moment model is trained on ~105 DFT-computed data points and predicts the dipole moment in Debye.
HOMO
HOMO provides a measure of the oxidative stability of a solvent. The HOMO model is trained on ~105 DFT-computed data points and predicts the highest occupied molecular orbital energy of a molecule in eV.
LUMO
LUMO provides a measure of the reductive stability of a solvent. The LUMO model is trained on ~105 DFT-computed data points and predicts the lowest occupied molecular orbital energy of a molecule in eV.
Input and Outputs
As part of the Google for Startups Cloud Program, Aionics is proud to serve these Molecular Property Prediction Models to the chemistry and materials communities through Google Cloud compute infrastructure and AI-focused products.