Granted Patent B2

US 12,378,610 · App. 18/506,690 · Granted Aug 5, 2025

Inventors: Christine Buerki (Vancouver, CA); Anamaria Crisan (Vancouver, CA); Elai Davicioni (La Jolla, CA); Nicholas George Erho (Vancouver, CA); Mercedeh Ghadessi (New Westminster, CA); Robert B. Jenkins (Rochester, MN); Ismael A. Vergara Correa (West Vancouver, CA)

Assignees: Veracyte SD, Inc.; Mayo Foundation for Medical Education and Research

G06N5/01

Abstract

Claims (40)

1. A system for facilitating cancer-related prediction accuracy of a trained model, without requiring renormalization of the entirety of a given training dataset for training the model for novel data, by arranging preprocessing vectors and a trained random forest model, that are derived from the same training dataset, to respectively preprocess sample target data and generate predictions with the preprocessed data, the system comprising:

one or more processors and non-transitory machine-readable media storing instructions that, when executed by the one or more processors, cause operations comprising:

accessing a random forest machine learning model comprising features that are derived from a training dataset and selected for the random forest machine learning model;

obtaining target data derived from a target sample;

normalizing, using frozen vectors derived from randomly-selected subsets of the training dataset, the target data to generate processed data; and

after generating the processed data using the frozen vectors, generating, using the random forest machine learning model on the processed data and without requiring renormalization of the entirety of the training dataset, a likelihood score related to cancer occurrence by inputting the processed data into nodes of the random forest machine learning model, wherein the nodes and the frozen vectors are both derived from the training dataset.

2. The system of claim 1 , wherein the features are selected for the random forest machine learning model based on the features having a non-zero co-efficient less than or equal to a p-value threshold in multiple bootstrapping processes performed on different subsets of the training dataset.

3. A method for facilitating cancer-related prediction accuracy of a trained model without requiring renormalization of the entirety of a given training dataset for training the model for novel data, the method comprising:

accessing a machine learning model comprising features that are derived from a training dataset and selected for the machine learning model;

obtaining target data derived from a target sample;

normalizing, using frozen vectors derived from randomly-selected subsets of the training dataset, the target data to generate processed data; and

after generating the processed data using the frozen vectors, generating, using the machine learning model on the processed data and without requiring renormalization of the entirety of the training dataset, a likelihood score related to cancer occurrence by inputting the processed data into nodes of the machine learning model, wherein the nodes and the frozen vectors are both derived from the training dataset.

4. The method of claim 3 , wherein the machine learning model comprises a random forest having decision trees.

5. The method of claim 3 , wherein the features have a non-zero co-efficient satisfying a predetermined threshold in multiple bootstrapping processes performed on different subsets of the training dataset.

6. The method of claim 3 , further comprising:

randomly selecting a plurality of subsets of the training dataset for the randomly-selected subsets for deriving the frozen vectors that are used to normalize the target data to generate the processed data.

7. The method of claim 3 , wherein:

obtaining the target data comprises obtaining one or more expression patterns derived from the target sample;

normalizing the target data comprises normalizing the one or more expression patterns to generate one or more processed expression patterns; and

generating the likelihood score comprises generating, using the machine learning model on the one or more processed expression patterns, the likelihood score related to the cancer occurrence.

8. The method of claim 3 , wherein generating the processed data comprises removing cross-hybridizing probes from the target data.

9. The method of claim 3 , wherein generating the likelihood score comprises generating, using the machine learning model on the processed data, a score indicating a probability of a stage of a plurality of stages of the cancer occurrence.

10. The method of claim 3 , wherein generating the likelihood score comprises generating, using the machine learning model on the processed data, a score indicating a probability of recurrence of the cancer occurrence.

11. One or more non-transitory machine-readable media storing instructions, that when executed by one or more processing devices, cause operations comprising:

accessing a machine learning model comprising features that are selected for the machine learning model and derived from a training dataset;

obtaining target data derived from a target sample;

generating, using frozen vectors derived from randomly-selected subsets of the training dataset, processed data from the target data; and

after generating the processed data using the frozen vectors, generating, using the machine learning model on the processed data and without requiring renormalization of the entirety of the training dataset, a likelihood score related to cancer occurrence by inputting the processed data into nodes of the machine learning model.

12. The media of claim 11 , wherein the machine learning model comprises a random forest having decision trees.

13. The media of claim 11 , wherein the target data is derived via an array on which probes specific to targets are attached.

14. The media of claim 11 , wherein each feature of the features of the machine learning model has a non-zero co-efficient greater than a predetermined threshold in multiple bootstrapping processes performed on different subsets of the training dataset.

15. The media of claim 11 , the operations further comprising:

randomly selecting a plurality of subsets of the training dataset for the randomly-selected subsets for deriving the frozen vectors that are used to normalize the target data to generate the processed data.

16. The media of claim 11 , wherein:

obtaining the target data comprises obtaining one or more expression patterns derived from the target sample;

normalizing the target data comprises normalizing the one or more expression patterns to generate one or more processed expression patterns; and

generating the likelihood score comprises generating, using the machine learning model on the one or more processed expression patterns, the likelihood score related to the cancer occurrence.

17. The media of claim 11 , wherein generating the processed data comprises removing cross-hybridizing probes from the target data.

18. The media of claim 11 , wherein generating the likelihood score comprises generating, using the machine learning model on the processed data, a score indicating a probability a stage of a plurality of stages of the cancer occurrence.

19. The media of claim 11 , wherein generating the likelihood score comprises generating, using the machine learning model on the processed data, a score indicating a probability of recurrence of the cancer occurrence.

Continuity (6)

References Cited (400)