Granted Patent B1

US 12,731,692 · App. 19/057,860 · Granted Sep 8, 2026

Inventors: Tathagata Banerjee (Waltham, MA); Karin Comer Knudson (Somerville, MA)

Assignee: Neumora Therapeutics, Inc.

G16H50/20G16H10/60

Abstract

Claims (80)

1 . A method performed by one or more computers for performing a computationally efficient simulation of a clinical trial using an ensemble of machine learning models and without requiring separate training of each machine learning model in the ensemble, the method comprising:

training, by the one or more computers, a probabilistic machine learning model to process an input feature set characterizing a subject to generate a predicted response score for the subject, wherein:

the predicted response score for the subject characterizes a predicted change in a medical condition of the subject during the clinical trial resulting from receiving a treatment; and

each model parameter in a set of model parameters of the probabilistic machine learning model is associated with a respective distribution over possible values of the model parameter; and

generating, by the one or more computers, the ensemble of machine learning models without requiring separate training of each machine learning model in the ensemble using the probabilistic machine learning model, comprising, for each machine learning model in the ensemble of machine learning models:

generating, by the one or more computers, the machine learning model as an instance of the probabilistic machine learning model by, for each model parameter of the probabilistic machine learning model, sampling a respective single value for the model parameter from the distribution over possible values of the model parameter;

performing, by the one or more computers, the simulation to predict an outcome of the clinical trial according to a set of inclusion criteria, comprising:

obtaining, by the one or more computers, data defining a simulated population of subjects satisfying the inclusion criteria for the clinical trial, wherein each subject in the simulated population of subjects is associated with a feature set characterizing the subject;

generating, by the one or more computers and using the ensemble of machine learning models, for each subject in the simulated population of subjects, a plurality of predicted response scores for the subject,

wherein for each subject, generating the plurality of predicted response scores for the subject comprises:

processing, by the one or more computers and using each machine learning model in the ensemble of machine learning models, the feature set characterizing the subject in accordance with values of a set of model parameters of the machine learning model to generate a respective predicted response score for the subject; and

determining, by the one or more computers, the predicted outcome of the clinical trial based on the plurality of predicted response scores for each subject in the simulated population of subjects; and

determining, by the one or more computers, a final set of inclusion criteria for the clinical trial based at least in part on the simulation of the clinical trial.

2 . The method of claim 1 , wherein each machine learning model in the ensemble of machine learning models has: (i) a same model architecture, and (ii) a respective set of model parameters having values that are specific to the machine learning model.

3 . The method of claim 1 , wherein the probabilistic machine learning model comprises one or more of a neural network model; or a linear regression model; or a decision tree model.

4 . The method of claim 1 , wherein obtaining data defining the simulated population of subjects satisfying the inclusion criteria for the clinical trial comprises:

generating a distribution over a space of feature sets based on the set of inclusion criteria for the clinical trial; and

generating the feature set for each subject in the simulated population of subjects by sampling from the distribution over the space of feature sets.

5 . The method of claim 4 , wherein generating the distribution over the space of feature sets based on the set of inclusion criteria for the clinical trial comprises:

identifying a real-world population of subjects that satisfy the set of inclusion criteria;

obtaining a collection of feature sets that includes a respective feature set for each subject in the real-world population of subjects that satisfy the inclusion criteria; and

generating the distribution over the space of feature sets by fitting the distribution over the space of feature sets to the collection of feature sets for the real-world population of subjects that satisfy the inclusion criteria.

6 . The method of claim 1 , wherein the clinical trial is a clinical trial for a drug; and

wherein the simulated population of subjects is a treatment group that receive the drug; and

wherein each predicted response score characterizes a predicted change in the medical condition of the subject during the clinical trial as a result of receiving the drug.

7 . The method of claim 6 , wherein determining the predicted outcome of the clinical trial for the drug based on the plurality of predicted response scores for each subject in the simulated population of subjects comprises:

determining a measure of central tendency of the predicted response scores for the subjects in the simulated population of subjects; and

determining a measure of dispersion of the predicted response scores for the subjects in the simulated population of subjects.

8 . The method of claim 6 , wherein the drug is a drug for treating depression; and

wherein each predicted response score for each subject characterizes a predicted change in a measure of depression of the subject during the clinical trial.

9 . The method of claim 6 , wherein performing the simulation to predict the outcome of the clinical trial having the set of inclusion criteria further comprises:

obtaining data defining a second simulated population of subjects satisfying the inclusion criteria for the clinical trial, wherein the second simulated population of subjects is a control group that does not receive the drug; and

generating, for each subject in the second simulated population of subjects, a plurality of predicted response scores for the subject;

wherein the predicted outcome of the clinical trial for the drug is based on both: (i) the plurality of predicted response scores for each subject in the simulated population of subjects that is the treatment group, and (ii) the plurality of predicted response scores for each subject in the second simulated population of subjects that is the control group.

10 . The method of claim 9 , wherein the predicted outcome of the clinical trial is based on a measure of difference between: (i) the plurality of predicted response scores for each subject in the simulated population of subjects that is the treatment group, and (ii) the plurality of predicted response scores for each subject in the second simulated population of subjects that is the control group.

11 . The method of claim 1 , wherein the clinical trial is for a drug, and further comprising:

selecting the set of inclusion criteria for the clinical trial based at least in part on the predicted outcome of the clinical trial having the inclusion criteria; and

for each of one or more real-world subjects:

determining that the real-world subject satisfies the set of inclusion criteria for the clinical trial; and

in response, determining that the real-world subject should receive the drug as part of the clinical trial.

12 . The method of claim 11 , further comprising, for each of the one or more real-world subjects:

administering the drug to the real-world subject in response to determining that the real-world subject should receive the drug as part of the clinical trial.

13 . The method of claim 1 , further comprising performing an automated search through a space of possible inclusion criteria to optimize a predicted outcome of the clinical trial.

14 . The method of claim 13 , wherein performing the automated search through the space of possible inclusion criteria to optimize the predicted outcome of the clinical trial comprises:

identifying a plurality of possible sets of inclusion criteria for the clinical trial;

performing, for each set of inclusion criteria in the plurality of possible sets of inclusion criteria, a respective simulation to predict an outcome of a clinical trial having the set of inclusion criteria; and

selecting a final set of inclusion criteria based at least in part on the predicted outcomes for the plurality of possible sets of inclusion criteria.

15 . The method of claim 13 , wherein the automated search through the space of possible inclusion criteria is performed in accordance with a black box optimization technique.

16 . The method of claim 13 , wherein the automated search through the space of possible inclusion criteria is subject to one or more constraints;

wherein the one or more constraints include a constraint that, for a set of inclusion criteria to be feasible, at least a number or percentage of real-world subjects are predicted to satisfy the inclusion criteria.

17 . A system comprising:

one or more computers; and

one or more storage devices communicatively coupled to the one or more computers, wherein the one or more storage devices store instructions that, when executed by the one or more computers, cause the one or more computers to perform operations for performing a computationally efficient simulation of a clinical trial using an ensemble of machine learning models and without requiring separate training of each machine learning model in the ensemble, the operations comprising:

training a probabilistic machine learning model to process an input feature set characterizing a subject to generate a predicted response score for the subject, wherein:

the predicted response score for the subject characterizes a predicted change in a medical condition of the subject during the clinical trial resulting from receiving a treatment; and

each model parameter in a set of model parameters of the probabilistic machine learning model is associated with a respective distribution over possible values of the model parameter; and

generating the ensemble of machine learning models without requiring separate training of each machine learning model in the ensemble using the probabilistic machine learning model, comprising, for each machine learning model in the ensemble of machine learning models:

generating the machine learning model as an instance of the probabilistic machine learning model by, for each model parameter of the probabilistic machine learning model, sampling a respective single value for the model parameter from the distribution over possible values of the model parameter;

performing the simulation to predict an outcome of the clinical trial according to a set of inclusion criteria, comprising:

obtaining data defining a simulated population of subjects satisfying the inclusion criteria for the clinical trial, wherein each subject in the simulated population of subjects is associated with a feature set characterizing the subject;

generating, using the ensemble of machine learning models and for each subject in the simulated population of subjects, a plurality of predicted response scores for the subject,

wherein for each subject, generating the plurality of predicted response scores for the subject comprises:

processing, by each machine learning model in the ensemble of machine learning models, the feature set characterizing the subject in accordance with values of a set of model parameters of the machine learning model to generate a respective predicted response score for the subject; and

determining the predicted outcome of the clinical trial based on the plurality of predicted response scores for each subject in the simulated population of subjects; and

determining a final set of inclusion criteria for the clinical trial based at least in part on the simulation of the clinical trial.

18 . One or more non-transitory computer storage media storing instructions that when executed by one or more computers cause the one or more computers to perform operations for performing a computationally efficient simulation of a clinical trial using an ensemble of machine learning models and without requiring separate training of each machine learning model in the ensemble, the operations comprising:

training a probabilistic machine learning model to process an input feature set characterizing a subject to generate a predicted response score for the subject, wherein:

the predicted response score for the subject characterizes a predicted change in a medical condition of the subject during the clinical trial resulting from receiving a treatment; and

each model parameter in a set of model parameters of the probabilistic machine learning model is associated with a respective distribution over possible values of the model parameter; and

generating the ensemble of machine learning models without requiring separate training of each machine learning model in the ensemble using the probabilistic machine learning model, comprising, for each machine learning model in the ensemble of machine learning models:

generating the machine learning model as an instance of the probabilistic machine learning model by, for each model parameter of the probabilistic machine learning model, sampling a respective single value for the model parameter from the distribution over possible values of the model parameter;

performing the simulation to predict an outcome of the clinical trial according to a set of inclusion criteria, comprising:

obtaining data defining a simulated population of subjects satisfying the inclusion criteria for the clinical trial, wherein each subject in the simulated population of subjects is associated with a feature set characterizing the subject;

generating, using the ensemble of machine learning models and for each subject in the simulated population of subjects, a plurality of predicted response scores for the subject,

wherein for each subject, generating the plurality of predicted response scores for the subject comprises:

processing, by each machine learning model in the ensemble of machine learning models, the feature set characterizing the subject in accordance with values of a set of model parameters of the machine learning model to generate a respective predicted response score for the subject; and

determining the predicted outcome of the clinical trial based on the plurality of predicted response scores for each subject in the simulated population of subjects; and

determining a final set of inclusion criteria for the clinical trial based at least in part on the simulation of the clinical trial.

19 . The non-transitory computer storage media of claim 18 , wherein each machine learning model in the ensemble of machine learning models has: (i) a same model architecture, and (ii) a respective set of model parameters having values that are specific to the machine learning model.

20 . The non-transitory computer storage media of claim 18 , wherein the probabilistic machine learning model comprises one or more of: a neural network model; or a linear regression model; or a decision tree model.

Assignments (1)

Continuity (1)

References Cited (9)