Depolymerization optimization platform
Computer-implemented methods may include accessing a predictive function. The predictive function may be configured to receive a partial or complete bond string and position (BSP) representation of a molecule of a reactant ionic liquid, where the representation identifies relative positions of atoms in the molecule. The predictive function may be configured to predict a reaction-characteristic value that characterizes a reaction between the ionic liquid and a particular polymer. The predictive function may be generated using training data corresponding to a set of molecules that were selected using Bayesian optimization, one or more previous versions of the predictive function, and experimentally derived reaction-characteristic values characterizing reactions between the molecules and the particular polymer. The method may also include identifying a particular ionic liquid as a prospect for depolymerizing the particular polymer based on the predictive function. The method may further include outputting an identification of the ionic liquid.
1 . A computer-implemented method comprising:
accessing a predictive function configured to:
receive a partial or complete bond string and position (BSP) representation of a molecule of a reactant ionic liquid, wherein the partial or complete BSP representation identifies relative positions of atoms connected by a bond in the molecule;
transform the partial or complete BSP representations into machine-learned molecular fingerprints by applying a graph neural network,
predict a reaction-characteristic value that characterizes a reaction between the reactant ionic liquid and a particular polymer based on the received partial or complete BSP representation,
wherein the predictive function was generated at least in part using:
training data corresponding to a set of molecules that were selected using Bayesian optimization, the training data being associated with the partial or complete BSP representations represented in a reduced dimensionality space by using one or more dimensionality reduction technique,
one or more previous versions of the predictive function,
experimentally derived reaction-characteristic values characterizing reactions between the molecules selected using Bayesian optimization and the particular polymer, and
the machine-learned molecular fingerprints of BSP molecular representation;
identifying, in the reduced dimensionality space, at least one particular ionic liquid as a prospect for depolymerizing the particular polymer based on the predictive function; and
outputting an identification of the at least one particular ionic liquid.
2 . The computer-implemented method of claim 1 , wherein the identification of the at least one particular ionic liquid includes a type of representation that is different from partial or complete BSP representation.
3 . The computer-implemented method of claim 1 , further comprising:
identifying a BSP representation for each of the molecules that were selected using Bayesian optimization, wherein the predictive function was generated at least in part based on the BSP representations of the molecules selected using Bayesian optimization.
4 . The computer-implemented method of claim 1 , further comprising:
generating the training data by a performing quantum-mechanical or thermodynamical method.
5 . The computer-implemented method of claim 1 , wherein the selection of the set of molecules corresponding to the training data was further based on evaluating a utility function using at least one of the one or more previous versions of the predictive function.
6 . The computer-implemented method of claim 1 , wherein the reaction-characteristic value includes a mixing enthalpy associated with the reactant ionic liquid and the particular polymer.
7 . The computer-implemented method of claim 1 , wherein the reaction-characteristic value includes an activity coefficient associated with the reactant ionic liquid and the particular polymer.
8 . A system comprising:
one or more data processors; and
a non-transitory computer readable storage medium containing instructions which, when executed on the one or more data processors, cause the one or more data processors to perform operations comprising a method comprising:
accessing a predictive function configured to:
receive a partial or complete bond string and position (BSP) representation of a molecule of a reactant ionic liquid, wherein the partial or complete BSP representation identifies relative positions of atoms connected by a bond in the molecule;
transform the partial or complete BSP representations into machine-learned molecular fingerprints by applying a graph neural network;
predict a reaction-characteristic value that characterizes a reaction between the reactant ionic liquid and a particular polymer based on the received partial or complete BSP representation,
wherein the predictive function was generated at least in part using:
training data corresponding to a set of molecules that were selected using Bayesian optimization, the training data being associated with the partial or complete BSP representations represented in a reduced dimensionality space by using one or more dimensionality reduction technique,
one or more previous versions of the predictive function,
experimentally derived reaction-characteristic values characterizing reactions between the molecules selected using Bayesian optimization and the particular polymer, and
the machine-learned molecular fingerprints of BSP molecular representation;
identifying, in the reduced dimensionality space, at least one particular ionic liquid as a prospect for depolymerizing the particular polymer based on the predictive function; and
outputting an identification of the at least one particular ionic liquid.
9 . The system of claim 8 , wherein the identification of the at least one particular ionic liquid includes a type of representation that is different from partial or complete BSP representation.
10 . The system of claim 8 , the method further comprising:
identifying a BSP representation for each of the molecules that were selected using Bayesian optimization, wherein the predictive function was generated at least in part based on the BSP representations of the molecules selected using Bayesian optimization.
11 . The system of claim 8 , the method further comprising:
generating the training data by a performing quantum-mechanical or thermodynamical method.
12 . The system of claim 8 , wherein the selection of the set of molecules corresponding to the training data was further based on evaluating a utility function using at least one of the one or more previous versions of the predictive function.
13 . The system of claim 8 , wherein the reaction-characteristic value includes a mixing enthalpy associated with the reactant ionic liquid and the particular polymer.
14 . The system of claim 8 , wherein the reaction-characteristic value includes an activity coefficient associated with the reactant ionic liquid and the particular polymer.
15 . A computer-program product tangibly embodied in a non-transitory machine-readable storage medium, including instructions configured to cause one or more data processors to perform a method comprising:
accessing a predictive function configured to:
receive a partial or complete bond string and position (BSP) representation of a molecule of a reactant ionic liquid, wherein the partial or complete BSP representation identifies relative positions of atoms connected by a bond in the molecule;
transform the partial or complete BSP representations into machine-learned molecular fingerprints by applying a graph neural network,
predict a reaction-characteristic value that characterizes a reaction between the reactant ionic liquid and a particular polymer based on the received partial or complete BSP representation;
wherein the predictive function was generated at least in part using:
training data corresponding to a set of molecules that were selected using Bayesian optimization, the training data being associated with the partial or complete BSP representations represented in a reduced dimensionality space by using one or more dimensionality reduction technique,
one or more previous versions of the predictive function,
experimentally derived reaction-characteristic values characterizing reactions between the molecules selected using Bayesian optimization and the particular polymer, and
the machine-learned molecular fingerprints of BSP molecular representation;
identifying, in the reduced dimensionality space, at least one particular ionic liquid as a prospect for depolymerizing the particular polymer based on the predictive function; and
outputting an identification of the at least one particular ionic liquid.
16 . The computer-program product of claim 15 , wherein the identification of the at least one particular ionic liquid includes a type of representation that is different from partial or complete BSP representation.
17 . The computer-program product of claim 15 , the method further comprising:
identifying a BSP representation for each of the molecules that were selected using Bayesian optimization, wherein the predictive function was generated at least in part based on the BSP representations of the molecules selected using Bayesian optimization.
18 . The computer-program product of claim 15 , the method further comprising:
generating the training data by a performing quantum-mechanical or thermodynamical method.
19 . The computer-program product of claim 15 , wherein the selection of the set of molecules corresponding to the training data was further based on evaluating a utility function using at least one of the one or more previous versions of the predictive function.
20 . The computer-program product of claim 15 , wherein the reaction-characteristic value includes a mixing enthalpy associated with the reactant ionic liquid and the particular polymer.