Gene expression analysis techniques using gene rankings and statistical models for identifying biological sample characteristics
Inventors
Antysheva, Zoia • Svekolkin, Viktor • Kotlov, Nikita • Karelin, Anton • Postovalova, Ekaterina
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-11482301-B2
Publication Date
2022-10-25
Expiration Date
Abstract
Techniques for determining one or more characteristics of a biological sample using rankings of gene expression levels in expression data obtained using one or more sequencing platforms is described. The techniques may include obtaining expression data for a biological sample of a subject. The techniques further include ranking genes in a set of genes based on their expression levels in the expression data to obtain a gene ranking and determining using the gene ranking and a statistical model, one or more characteristics of the biological sample.
Core Innovation
The invention provides a system and method that use a trained statistical model to determine at least one characteristic of a biological sample in a subject having, suspected of having or at risk of having cancer. Expression data for a biological sample is obtained at least in part by sequencing and includes first expression levels for a set of genes, and at least some genes are ranked based on the first expression levels to obtain a gene ranking including values identifying relative ranks that are different from the first expression levels.
The trained statistical model comprises parameters estimated using training data indicating a plurality of gene rankings of at least some genes in a set of genes, where the parameters are estimated after the at least some genes in the set of genes were selected. The gene ranking is provided as an input to the trained statistical model and processed using the parameters and the gene ranking to obtain an output indicating the at least one characteristic, where the characteristic is a physiological characteristic of cells in the biological sample or a physiological characteristic of a tissue from which the cells originate.
In described examples, iterative gene-set selection uses ranked gene expression and statistical model hyperparameter fitting with k-fold cross-validation scoring to choose a high-performing gene set under a minimum-size stopping criterion based on quality score improvement while reducing the number of genes. The document further describes classifier-based implementations for cancer-related characteristics using curated gene sets, including DLBCL cell of origin using a gradient-boosted decision tree classifier and SHAP-based feature importance, and HPV status prediction in head and neck squamous cell carcinoma using curated gene sets.
Additional classification configurations are described for PTCL subtype classification using gene ranking inputs with multi-class or multiple/binary classifier configurations, including representative gene lists from named tables. Across these examples, the invention frames gene ranking as an input representation distinct from the original expression levels, and uses the trained statistical model to output physiological or cell/tissue-origin-related characteristics associated with cancer.
Claims Coverage
The document includes three independent claim types: a system, a method, and a non-transitory computer-readable storage medium. Across them, the inventive coverage centers on using a trained statistical model whose parameters are estimated after gene selection, ranking genes from sequencing-derived expression levels into relative-rank values different from the expression levels, and feeding the gene ranking into the trained statistical model to output a physiological characteristic of cells or tissue origin.
Gene ranking from expression levels for trained model input
ranking the at least some genes in the set of genes based on their first expression levels in the expression data to obtain a gene ranking, the gene ranking including values identifying relative ranks of the at least some genes in the gene ranking, wherein the values are different from the first expression levels
Trained statistical model with parameters estimated after gene selection
accessing a trained statistical model comprising parameters estimated using training data, the training data indicating a plurality of gene rankings of at least some genes in a set of genes, the parameters being estimated after the at least some genes in the set of genes were selected
Determine physiological characteristic by processing gene ranking output
determining at least one characteristic of the biological sample by providing the gene ranking as an input to the trained statistical model and processing the gene ranking with the trained statistical model to obtain an output indicating the at least one characteristic of the biological sample, the processing of the gene ranking with the trained statistical model comprises using the parameters and the gene ranking provided as the input to obtain the output indicating the at least one characteristic of the biological sample
Physiological characteristic of cells or tissue of origin
the at least one characteristic of the biological sample is a physiological characteristic of cells in the biological sample or a physiological characteristic of a tissue from which the cells in the biological sample originate
Non-transitory medium for the same gene-ranking/statistical-model pipeline
storing processor-executable instructions that are configured to cause, when executed by at least one computer hardware processor, the at least one computer hardware processor to perform: accessing a trained statistical model comprising parameters estimated using training data; obtaining expression data previously obtained at least in part by sequencing; ranking genes to obtain a gene ranking including values identifying relative ranks different from the first expression levels; and determining at least one characteristic by providing the gene ranking to the trained statistical model
Output characteristic includes at least one cancer-related physiological descriptor (via dependents)
the at least one characteristic of the biological sample is a physiological characteristic of cells in the biological sample or a physiological characteristic of a tissue from which the cells in the biological sample originate (with dependent refinements specifying cancer grade, tissue of origin/tissue type, cancer subtype, and viral status including HPV status)
Across the independent claims, the inventive coverage centers on using a trained statistical model whose parameters are estimated from training data consisting of gene rankings of selected genes, where the system or method ranks genes from sequencing-derived expression levels into relative-rank values different from the first expression levels, and then determines a physiological characteristic of cells or tissue origin by processing the gene ranking with the trained statistical model.
Stated Advantages
Iterative gene-set selection uses a minimum-size stopping criterion based on quality score improvement while reducing the number of genes.
Documented Applications
DLBCL cell of origin using a gradient-boosted decision tree classifier and SHAP-based feature importance.
HPV status prediction in head and neck squamous cell carcinoma using curated gene sets.
PTCL subtype classification using gene ranking inputs with multi-class or multiple/binary classifier configurations.
Interested in licensing this patent?