Matrix factorization of antibiogram metadata

Patent metadata

Field	Value
Publication number	US-11646117-B2
Application number	US-201916430856-A
Country	US
Kind code	B2
Filing date	Jun 4, 2019
Priority date	Jun 4, 2019
Publication date	May 9, 2023
Grant date	May 9, 2023

Abstract

A method is described that utilizes non-negative matrix factorization to predict susceptibility of a microorganism to an antimicrobial drug. A sparse adjacency matrix is constructed from existing ground truth datasets that include antibiogram data and other data associated with microorganisms. The rows of the adjacency matrix correspond to biosamples, and the columns correspond to instances of metadata and drugs associated with one or more of the biosamples. The elements of the adjacency matrix are assigned non-zero numerical values or zero depending on whether a known association exists. The adjacency matrix is then factored using a selected number of latent factors, thereby producing a reconstruction matrix approximating the adjacency matrix. The values of the reconstruction matrix are used to predict antimicrobial susceptibility of a biosample ID to a drug when antibiogram data are lacking.

First claim

What is claimed is:

1. A method, comprising: providing an initial matrix comprising rows corresponding to microorganisms and columns corresponding to metadata associated with one or more of the microorganisms, the metadata including antibiogram data, wherein an element of the initial matrix linking a microorganism to an antibiotic drug has a numerical value indicating no antibiogram data exist for the microorganism with respect to the antibiotic drug; factoring the initial matrix using a matrix factorization algorithm, thereby forming a first factor matrix and a second factor matrix, wherein training data is used to train the matrix factorization algorithm, the training data comprising bacteria isolate-antibiotic pairs, generating the training data comprising: obtaining ground truth data for the microorganism, extracting the metadata from the ground truth data, and cleaning the ground truth data by transforming the metadata having been extracted into discrete data according to a type of the matrix factorization algorithm, wherein the matrix factorization algorithm learns latent factors between the first factor matrix and the second factor matrix; multiplying the first factor matrix by the second factor matrix, thereby forming a reconstruction matrix, wherein an element of the reconstruction matrix linking the microorganism to the antibiotic drug has a numerical value indicating the microorganism is resistant or susceptible to the antibiotic drug; predicting antibiotic resistance of the microorganism to the antibiotic drug according to the reconstruction matrix; and selecting the antibiotic drug to treat the microorganism in a medical therapy based on the numerical value obtained from the reconstruction matrix.

2. The method of claim 1 , wherein said factoring the initial matrix is performed using k number of the latent factors, wherein k is a positive whole number greater than 1.

3. The method of claim 2 , wherein k is a positive whole number between 1 and 1,000,000.

4. The method of claim 2 , wherein: the initial matrix has m rows and n columns, m and n being positive integers greater than 1, the first factor matrix has m rows and k columns, and the second factor matrix has k rows and n columns; and the matrix factorization algorithm comprises weighted approximate rank pairwise (WARP) loss.

5. The method of claim 2 , wherein the k number of the latent factors are determined by an iterative process comprising i) choosing an initial number of latent factors k′, ii) factoring the initial matrix into a first factor matrix and a second factor matrix, iii) calculating a reconstruction matrix by multiplying the first factor matrix times the second factor matrix, iv) determining a reconstruction error between the initial matrix and the reconstruction matrix, and v) repeating i)-iv) until the reconstruction error is acceptable, wherein the initial number of latent factors is equal to k.

6. The method of claim 1 , wherein each entry of the first factor matrix and the second factor matrix is zero or a positive number.

7. The method of claim 1 , wherein at least one column of the initial matrix is assigned a taxonomic level.

8. The method of claim 1 , wherein each unique antibiotic drug of the metadata is assigned a different column of the initial matrix.

9. The method of claim 1 , wherein at least one column of the initial matrix is assigned to an antibiotic drug used in an antibiogram.

10. The method of claim 1 , wherein each of the microorganisms of the initial matrix is a member of the group consisting of bacteria, fungi, viruses, protozoans, and parasites.

11. The method of claim 1 , wherein at least one column of the initial matrix is assigned to an isolation source of the microorganisms.

12. A computer program product, comprising a computer readable hardware storage device having a computer-readable program code stored therein, said program code configured to be executed by a processor of a computer system to implement a method comprising: providing an initial matrix comprising rows corresponding to microorganisms and columns corresponding to metadata associated with one or more of the microorganisms, the metadata including antibiogram data, wherein an element of the initial matrix linking a microorganism to an antibiotic drug has a numerical value indicating no antibiogram data exist for the microorganism with respect to the antibiotic drug; factoring the initial matrix using a matrix factorization algorithm, thereby forming a first factor matrix and a second factor matrix, wherein training data is used to train the matrix factorization algorithm, the training data comprising bacteria isolate-antibiotic pairs, generating the training data comprising: obtaining ground truth data for the microorganism, extracting the metadata from the ground truth data, and cleaning the ground truth data by transforming the metadata having been extracted into discrete data according to a type of the matrix factorization algorithm, wherein the matrix factorization algorithm learns latent factors between the first factor matrix and the second factor matrix; multiplying the first factor matrix by the second factor matrix, thereby forming a reconstruction matrix, wherein an element of the reconstruction matrix linking the microorganism to the antibiotic drug has a numerical value indicating the microorganism is resistant or susceptible to the antibiotic drug; and selecting the antibiotic drug to treat the microorganism in a medical therapy based on the numerical value.

13. The computer product of claim 12 , wherein the computer program product identifies differences between entries of the reconstruction matrix and corresponding entries of the initial matrix and provides a report of the identified differences to a user.

14. A system comprising one or more computer processor circuits configured and arranged to: provide an initial matrix comprising rows corresponding to microorganisms and columns corresponding to metadata associated with one or more of the microorganisms, the metadata including antibiogram data, wherein an element of the initial matrix linking a microorganism to an antibiotic drug has a numerical value indicating no antibiogram data exist for the microorganism with respect to the antibiotic drug; factor the initial matrix using a matrix factorization algorithm, thereby forming a first factor matrix and a second factor matrix, wherein training data is used to train the matrix factorization algorithm, the training data comprising bacteria isolate-antibiotic pairs, generating the training data comprising: obtaining ground truth data for the microorganism, extracting the metadata from the ground truth data, and cleaning the ground truth data by transforming the metadata having been extracted into discrete data according to a type of the matrix factorization algorithm, wherein the matrix factorization algorithm being configured to learn latent factors between the first factor matrix and the second factor matrix; multiply the first factor matrix by the second factor matrix, thereby forming a reconstruction matrix, wherein an element of the reconstruction matrix linking the microorganism to the antibiotic drug has a numerical value indicating the microorganism is resistant or susceptible to the antibiotic drug; and selecting the antibiotic drug to treat the microorganism in a medical therapy based on the numerical value.

15. The system of claim 14 , wherein the system is located at a cloud platform.

16. A method, comprising: acquiring antibiogram data that include a table in which the rows correspond to biological samples and the columns correspon

Assignees

• IBM

Inventors

• Kunitomi Mark
• Zubarev Dmitry
• Swaminathan Sarathkrishna

Classifications

• G16H70/60

  relating to pathologies · CPC title

• G16B40/30

  Unsupervised data analysis · CPC title

• G16H50/30

  for calculating health indices; for individual health risk assessment · CPC title

• G16B20/00

  ICT specially adapted for functional genomics or proteomics, e.g. genotype-phenotype associations · CPC title

• G06N7/00

  Computing arrangements based on specific mathematical models · CPC title