Methods and compositions for classification of samples
Inventors
Wilde, Jonathan I. • Chudova, Darya • Kennedy, Giulia C.
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-10446272-B2
Publication Date
2019-10-15
Expiration Date
Abstract
Disclosed herein are kits, compositions, and methods relating to the classification of samples. Methods disclosed herein can be used to identify sample mix-ups. Methods disclosed herein can also be used to diagnose conditions or to support treatment-related decisions.
Core Innovation
A method is described for identifying a subject as having a cancer by combining cytological testing with gene-expression profiling and a trained machine learning classifier. A first portion of a sample is subjected to cytological testing when the sample is ambiguous or suspicious, and a second portion is assayed for levels of expression of a set of genes that does not include a BRAF gene.
The measured expression levels are used as inputs for applying a trained machine learning classifier, and one or more technical factor variables are removed upon classifying the sample. The method outputs an electronic report that classifies the sample as containing or not containing a BRAF mutation and identifies the subject as having the cancer based on the gene expression levels.
In the clinical context described, a thyroid cancer gene expression classifier is applied to indeterminate FNA cytology categories, including AUS/FLUS and FN/SFN. The system is described in the context of using gene-expression biomarker panels and trained algorithms to support diagnostic reporting and related molecular interpretation.
Claims Coverage
One independent claim framework is presented, combining cytological ambiguity or suspicion with gene-expression assaying that excludes BRAF genes, application of a trained machine learning classifier with technical factor variable removal, and output of an electronic report identifying a subject as having cancer based on BRAF mutation classification from gene expression levels.
Cytology-guided transition to BRAF-free gene-expression assaying
A first portion of a sample is subjected to cytological testing indicating the sample is ambiguous or suspicious, and then a second portion is assayed for expression levels of a set of genes, wherein the set of genes does not include a BRAF gene.
Trained machine learning classification with technical factor variable removal
A trained machine learning classifier is applied to the levels of expression to classify the sample as containing or not containing a BRAF mutation, wherein one or more technical factor variables are removed upon classifying the sample.
Electronic reporting of BRAF mutation status and cancer identification
An electronic report is output that classifies the sample as containing or not containing the BRAF mutation and identifies the subject as having the cancer based on the levels of expression.
The claims center on a workflow that uses cytological ambiguity to trigger expression profiling of a BRAF-excluding gene set, feeds the expression levels into a trained machine learning classifier that removes technical factor variables, and outputs an electronic report for both BRAF mutation presence and cancer identification.
Stated Advantages
Benign classifier results support watchful waiting versus surgery.
Improved negative predictive value, including prevalence-adjusted NPV reported as greater than 95%.
Broad validation across more than 4000 samples and more than 12 thyroid neoplasm subtypes.
Potential cost and QALY benefits via reduced resection rates.
Documented Applications
Applying a thyroid cancer gene expression classifier to indeterminate FNA cytology categories (AUS/FLUS, FN/SFN) to support management decisions such as watchful waiting versus surgery.
Interested in licensing this patent?