Materials information database including machine learning models
Examples are disclosed that relate to materials discovery using machine learning models. One example provides a method enacted on a computing system. The method comprises receiving a query comprising one or more of element information and material property information, and, based on the query, retrieving material data from a materials information database. The material data comprises structural information for each material within a set of materials matching the query, the set comprising one or more materials, and for one or more materials in the set of materials, one or more predicted material properties determined using one or more trained machine learning models. The method further comprises outputting the material data.
1 . A computing system, comprising:
a logic subsystem; and
a storage subsystem holding instructions executable by the logic subsystem to:
receive a query comprising one or more of element information and material property information,
based on the query, retrieve material data from a materials information database, the material data comprising,
structural information for each material within a set of materials matching the query, the set comprising one or more materials, and
for one or more materials in the set of materials, one or more predicted material properties determined using one or more trained machine learning models, and
output the material data.
2 . The computing system of claim 1 , wherein the instructions are further executable to
determine that an uncertainty for a predicted material property for a selected material is greater than an uncertainty threshold,
in response, perform a quantum mechanics (QM) calculation for the selected material to determine a QM-calculated value of a selected material property of the selected material, and
output the QM-calculated value of the selected material property.
3 . The computing system of claim 2 , wherein each predicted material property is predicted using a corresponding trained machine learning model, and wherein the instructions are further executable to use the QM-calculated value of the selected material property to further train the corresponding trained machine learning model for the selected predicted material property.
4 . The computing system of claim 1 , wherein the instructions are further executable to, for a material matching the query, determine the structural information by inputting a candidate structure into a trained machine learning model configured to perform structural relaxation and output a relaxed candidate structure.
5 . The computing system of claim 1 , wherein the predicted material property comprises one of a predicted band gap, a predicted conductivity, a predicted bulk modulus, a predicted shear modulus, a predicted formation energy, a predicted phonon density of states peak, a dielectric constant, or a refractive index.
6 . The computing system of claim 1 , wherein the one or more trained machine learning model comprises one or more graph neural networks.
7 . The computing system of claim 1 , wherein the material data comprises an uncertainty for a predicted material property of the one or more predicted material properties.
8 . The computing system of claim 7 , wherein the uncertainty for the predicted material property is determined by obtaining an ensemble of predicted values for the predicted material property from an ensemble of respective machine learning models, and estimating the uncertainty based on the ensemble of predicted values.
9 . The computing system of claim 1 , wherein the instructions are executable to input the query into a large language model configured to extract the one or more of element information and material property information.
10 . A method enacted on a computing system, the method comprising:
receiving a query comprising element information;
inputting the element information into a trained generative machine learning model to generate a plurality of candidate structures based on the element information; and
for one or more candidate structures of the plurality of candidate structures
inputting the candidate structure into a second trained machine learning model configured to output a predicted material property, and
outputting the candidate structure and the predicted material property.
11 . The method of claim 10 , further comprising determining that a materials information database does not contain information for any material matching the query prior to inputting the element information into the trained generative machine learning model.
12 . The method of claim 11 , further comprising updating the materials information database to include the candidate structure and the predicted material property for the candidate structure.
13 . The method of claim 10 , wherein the second trained machine learning model comprises an ensemble of trained machine learning models, and the method comprises
obtaining an ensemble of predicted values for the predicted material property from the ensemble of machine learning models, and
estimating the uncertainty based on the ensemble of predicted values.
14 . The method of claim 10 , further comprising
performing a quantum mechanics (QM) calculation of a selected material to determine a QM-calculated value of a selected material property, and
outputting the QM-calculated value of the selected material property.
15 . The method of claim 14 , wherein each predicted material property is predicted using a corresponding trained machine learning model, and further comprising using the QM-calculated value of the selected material property to further train the corresponding trained machine learning model for the selected predicted material property.
16 . A computing system, comprising:
a logic subsystem; and
a storage subsystem holding instructions executable by the logic subsystem to:
receive a query comprising element information,
input the element information into a trained generative machine learning model to generate a plurality of candidate structures based on the element information; and
for one or more candidate structures of the plurality of candidate structures
input the candidate structure into a second trained machine learning model configured to output a predicted material property, and
output the candidate structure and the predicted material property.
17 . The computing system of claim 16 , wherein the instructions are further executable to determine that a materials information database does not contain information for any material matching the query prior to inputting the element information into the trained generative machine learning model.
18 . The computing system of claim 17 , wherein the instructions are further executable to update the materials information database to include the candidate structure and the predicted material property for the candidate structure.
19 . The computing system of claim 16 , wherein the instructions are further executable to input the query into a large language model configured to extract the element information.
20 . The computing system of claim 16 , wherein the instructions are further executable to
receive a user input requesting to perform a QM calculation for the candidate structure to determine a QM-calculated relaxed candidate structure and a QM-calculated value of a selected material property,
form a material record comprising structural information based on the QM-calculated relaxed candidate structure and the QM-calculated value of the selected material property, and
output the material record for storage in an account of the user.