Algorithms for disease diagnostics
Inventors
Kennedy, Giulia C. • Chudova, Darya I. • Wang, Eric T. • Wilde, Jonathan I.
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-10731223-B2
Publication Date
2020-08-04
Expiration Date
Abstract
The present invention relates to compositions and methods for molecular profiling and diagnostics for genetic disorders and cancer, including but not limited to gene expression product markers associated with cancer or genetic disorders. In particular, the present invention provides algorithms and methods of classifying cancer, for example, thyroid cancer, methods of determining molecular profiles, and methods of analyzing results to provide a diagnosis.
Core Innovation
The invention provides algorithms for molecular profiling diagnostics that output an electronic report identifying a classification of a sample as benign or non-benign for lung cancer. The workflow begins with a portion of a subject sample that is cytologically tested and identified as ambiguous. After identifying ambiguity, one or more gene expression products are assayed from a second portion of the sample to generate a data set including data corresponding to a level of the gene expression products.
The generated data set is input into a trained algorithm to generate a classification of the sample as benign or non-benign for the lung cancer. The trained algorithm is trained with a training set of samples comprising known benign samples and known non-benign samples. The document describes using data-processing and model-building approaches that address variability and support decision-making, including technical-factor removal and normalization/adjustment of gene expression product data.
The described approaches include selecting biomarker panels using cross-validation and repeatability-based feature selection, including repeatability-based selection and Bayesian marker selection and combination approaches. The document further describes classifying samples using machine-learning classifiers and generating confidence-related performance summaries, including ROC curves and metrics such as PPV and NPV, to support the output report.
The background and described feasibility/performance include molecular profiling using gene expression and alternative splicing or exon usage, as well as adequacy/QC measurements. The document describes sample types and QC outcomes such as AUC, pDET, and RIN, and reports experimental feasibility across discovery datasets and classifier training with known benign/malignant labels in the context of ambiguous cytology.
Claims Coverage
The partial content provides one independent claim. It includes a trained algorithm for classifying a lung-cancer sample as benign or non-benign after ambiguous cytological testing, using gene expression product data from a second sample portion and electronically outputting a report; an associated dependent set refines constraints such as technical-factor handling and outcome granularity.
Ambiguous cytology triggered gene expression product profiling
identifying the sample as ambiguous by cytological testing and, after identifying ambiguity, assaying one or more gene expression products from a second portion of the sample to generate data corresponding to a level of the gene expression products
Trained algorithm producing benign versus non-benign lung-cancer classification
inputting the generated data into a trained algorithm that is trained with a training set of samples comprising known benign samples and known non-benign samples to generate a classification of the sample as benign or non-benign for lung cancer
Electronic reporting of benign versus non-benign classification
electronically outputting a report that identifies the classification of the sample as benign or non-benign for lung cancer
Across the independent claim, the core inventive coverage centers on a workflow that uses ambiguous cytological testing to trigger gene expression product assaying from a second portion, followed by classification using a trained algorithm trained on known benign and known non-benign samples, and electronic reporting of the benign versus non-benign result.
Stated Advantages
Provides an electronically output report identifying a classification of a sample as benign or non-benign for lung cancer.
Supports classification decisions using a trained algorithm based on known benign and known non-benign training samples.
Describes performance/confidence evaluation using ROC and metrics including PPV and NPV.
Documented Applications
Classifying a subject lung-cancer sample as benign or non-benign when cytological testing identifies the sample as ambiguous, with an electronically output report.
Interested in licensing this patent?