Functional deep neural network for high-dimensional data analysis

Patent metadata

Field	Value
Publication number	US-12080426-B2
Application number	US-202117217082-A
Country	US
Kind code	B2
Filing date	Mar 30, 2021
Priority date	Apr 6, 2020
Publication date	Sep 3, 2024
Grant date	Sep 3, 2024

Abstract

Various examples of methods and systems are provided related to functional deep neural networks (FDNNs), which can be used for high dimensional data analysis. In one example, a FDNN can be trained with a training set of omic data to produce a trained FDNN model. The likelihood of a condition can be determined based upon output indications of the FDNN corresponding to the one or more phenotypes, with the output indications based upon analysis of omic data including a multi-level omic profile from an individual by the trained FDNN. The FDNN model can include a series of basis functions as layers to capture complexity between the omic data with disease phenotypes. A treatment or prevention strategy for the individual can be identified based at least in part upon the likelihood of the condition.

First claim

Therefore, at least the following is claimed:

1. A method for risk prediction using high-dimensional omic data, comprising:
training, by at least one computing device, a functional deep neural network (FDNN) with a training set of omic data to produce a trained FDNN model, the trained FDNN model comprising a series of basis functions as a plurality of layers that capture complexity between the omic data with disease phenotypes after training, the training set of omic data comprising biomarkers applied as inputs to the FDNN and one or more phenotypes, the trained FDNN model configured to provide output indications in a standardized format;
obtaining, by the at least one computing device, omic data corresponding to an individual, the omic data comprising a multi-level omic profile from the individual stored in a database;
determining, by the at least one computing device, a likelihood of a condition associated with a disease based upon output indications of the FDNN corresponding to the one or more phenotypes, the output indications based upon analysis of the omic data comprising the multi-level omic profile from the individual by the trained FDNN;
identifying, by the at least one computing device, a treatment for the individual based at least in part upon the likelihood of the condition; and
generating, by the at least one computing device, an indication of the treatment for rendering on a networked computing device, the indication of treatment transmitted to all authorized medical or clinical personnel over a network for real time access via networked computing devices.

2. The method of claim 1 , wherein a first layer of the plurality of layers comprises a univariate function and remaining layers of the plurality of layers comprise a bivariate function.

3. The method of claim 1 , wherein the training set of omic data comprises risk predictors related to the one or more phenotypes, the risk predictors including biomarkers or established risk predictors.

4. The method of claim 3 , wherein the one or more phenotypes comprise disease diagnostic assessments, multiple correlated phenotypes, or high-dimensional phenotypes.

5. The method of claim 4 , wherein the high-dimensional phenotypes comprise biomarkers or neuroimaging data.

6. The method of claim 1 , wherein the plurality of layers of the FDNN are built via functional linear models with functional coefficients as weights in individual layers.

7. The method of claim 6 , wherein the plurality of layers of the FDNN adopts a penalty on a second-order derivative of the basis functions to ensure smoothness of the basis functions.

8. The method of claim 1 , wherein weights and biases in the FDNN are functions, and the FDNN takes an integral of functional coefficients in individual layers.

9. A system for risk prediction using high-dimensional omic data, comprising:
at least one computing device comprising processing circuitry including a processor and memory; and
a FDNN analysis program that, when executed by the processing circuitry, cause the at least one computing device to:
obtain an omic profile of an individual, the omic profile comprising a multi-level omic profile from the individual stored in a database;
determining a likelihood of a condition associated with a disease based upon output indications of a functional deep neural network (FDNN) corresponding to one or more phenotypes, the output indications based upon analysis of omic data comprising the multi-level omic profile by the FDNN, where the FDNN was trained with a training set of omic data to produce a trained FDNN model, the trained FDNN model comprising a series of basis functions as a plurality of layers that capture complexity between the omic data with disease phenotypes, the trained FDNN model configured to provide the output indications in a standardized format;
providing a treatment identified for the individual based at least in part upon the likelihood of the condition; and
generating an indication of the treatment for rendering on a networked computing device or another computing device, the indication of treatment transmitted to all authorized medical or clinical personnel over a network for real time access via networked computing devices.

10. The system of claim 9 , wherein the training set of omic data comprising biomarkers applied as inputs to the FDNN and the one or more phenotypes.

11. The system of claim 10 , wherein the training set of omic data comprises risk predictors related to the one or more phenotypes, the risk predictors including biomarkers or established risk predictors.

12. The system of claim 11 , wherein the one or more phenotypes comprise disease diagnostic assessments, multiple correlated phenotypes, or high- dimensional phenotypes.

13. The system of claim 12 , wherein the high-dimensional phenotypes comprise multi-level omic or neuroimaging data.

14. The system of claim 9 , wherein a first layer of the plurality of layers comprises a univariate function and remaining layers of the plurality of layers comprise a bivariate function.

15. The system of claim 9 , wherein the plurality of layers of the FDNN are built via functional linear models with functional coefficients as weights in individual layers.

16. The system of claim 15 , wherein the plurality of layers of the FDNN adopts a penalty on a second-order derivative of the basis functions to ensure smoothness of the basis functions.

17. The system of claim 9 , wherein weights and biases in the FDNN are functions, and the FDNN takes an integral of functional coefficients in individual layers.

Assignees

• Univ Florida
• Univ Florida

Inventors

• Lu Qing
• Zhang Shan
• Hou Tingting

Classifications

• G06N3/0499

  Feedforward networks · CPC title

• G06N3/09

  Supervised learning · CPC title

• G16H30/40

  for processing medical images, e.g. editing · CPC title

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title

• G06T2207/20081

  Training; Learning · CPC title