Classifier generation methods and predictive test for ovarian cancer patient prognosis under platinum chemotherapy
Inventors
Steingrimsson, Arni • Röder, Joanna • Grigorieva, Julia • Röder, Heinrich • Meyer, Krista
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-11621057-B2
Publication Date
2023-04-04
Expiration Date
Abstract
A method of generating a classifier includes a step of classifying each member of a development set of samples with a class label in a binary classification scheme with a first classifier; and generating a second classifier using a classifier development process with an input classifier development set being the members of the development set assigned one of the two class labels in the binary classification scheme by the first classifier. The second classifier stratifies the members of the set with an early label into two further sub-groups. We also describe identifying a plurality of different clinical sub-groups within the development set based on the clinical data and for each of the different clinical sub-groups, conducting a classifier generation process for each of the clinical sub-groups thereby generating clinical subgroup classifiers. We further describe an example of a hierarchical arrangement of such classifiers and their use in predicting, in advance of treatment, ovarian cancer patient outcomes on platinum-based chemotherapy.
Core Innovation
The invention provides a classifier predicting in advance whether an ovarian cancer patient is likely to be platinum-refractory or platinum-resistant when treated with platinum-based chemotherapy, using blood-based mass spectral data. The classifier uses mass spectral data stored as class-labeled reference sets, where the mass spectral data are included in a feature table of intensity values of a multitude of mass spectral features. The reference set includes class labels including an Early class label indicating relatively poor performance and a Late class label indicating relatively good performance.
The classifier performs hierarchical multi-level classification in series using a hierarchical structure generated at least from a master classifier. A plurality of k-nearest neighbor mini-classifiers are generated using identified sets of feature values in the mass spectral features, and for each mini-classifier a set of proposed classifications is derived for each sample in the reference set. A subset of mini-classifiers is identified with a threshold number of proposed classifications that correspond with class labels in the reference set, and the mini-classifiers in the subset are combined to generate a master classifier.
The hierarchical multi-level classification assigns class labels across at least first and second levels, and further levels may be added. At the first level, the classification algorithm produces an Early class label or a Late class label, where the Late class label identifies patients as likely to not be platinum-refractory or platinum-resistant and the Early class label triggers a second level for further stratification. The second level identifies patients as likely to be platinum-refractory or platinum-resistant when the first level indicates Early, and additional stratification including exceptionally poor and exceptionally good prognoses is provided through further levels and subgroup-specific classifiers.
Claims Coverage
The partial content includes three independent claims (clm-00001, clm-00007, clm-00011). Across these independent claims, the inventive subject matter centers on hierarchical multi-level classification using k-nearest neighbor mini-classifiers combined into master classifiers, using blood-based mass spectral feature tables with Early/Late class labels for ovarian cancer patients treated with platinum-based chemotherapy, and optional subgroup-based multi-stage extensions.
Early/Late labeled reference-set mass spectral classification for platinum-refractory/platinum-resistant prediction
A programmed computer storing a reference set of class-labeled mass spectral data obtained from blood-based samples of other ovarian cancer patients treated with platinum-based chemotherapy, where the mass spectral data are included in a feature table of intensity values of a multitude of mass spectral features and include an Early class label and a Late class label, and comparing mass spectral data of a sample to be tested with the reference set to generate a class label for the sample to be tested.
Hierarchical multi-level classification using kNN mini-classifiers and master classifier
A hierarchical multi-level classification generated by generating a plurality of mini-classifiers using a k-nearest neighbor (kNN) classification algorithm, deriving proposed classifications for each sample in the reference set for each mini-classifier, identifying a subset of mini-classifiers with a threshold number of proposed classifications that correspond with class labels, and combining the subset to generate a master classifier used to produce at least the hierarchical multi-level classification.
First level Early/Late with second level stratification into platinum-refractory/platinum-resistant
A first level producing either the Early class label or the Late class label, where the Late class label identifies patients as likely to not be platinum-refractory or platinum-resistant, and responsive to the first level producing the Early class label, proceeding to a second level that uses a subset of the reference set with the Early class label further stratified into an Earlier class label and a Later class label to identify patients as likely to be platinum-refractory or platinum-resistant.
Multi-stage classifier with early/late groups and exceptionally poor/good prognosis stages
A multi-stage classifier in which a first stage classifier stratifies test mass spectral data into either an Early or Late group; a second stage classifier further stratifies the Early group into Early and Late groups or Earlier and Later groups, implemented if the first stage classifies into the Early group and the Early class label produced by the second stage is associated with an exceptionally poor prognosis; and a third stage classifier further stratifies the Late group into Early and Late groups or Earlier and Later groups, implemented if the first stage classifies into the Late group and the Late class label produced by the third stage is associated with an exceptionally good prognosis.
Clinical-subgroup development generating C1…CN classifiers combined into a hierarchical multi-level classifier
A method of generating a classifier for classifying a test sample using a development set where dividing the development set into different clinical subgroups 1…N (N≥2), performing a classifier development process for each clinical subgroup to generate different classifiers C1…CN, generating for each clinical subgroup a plurality of kNN mini-classifiers with identified sets of feature values, deriving proposed classifications, identifying a subset of mini-classifiers with a threshold number of proposed classifications corresponding with class labels, combining the subset into a master classifier per subgroup, and combining the master classifiers generated for each clinical subgroup as part of a hierarchical multi-level classifier for a final classification process.
Across the independent claims, the coverage is focused on using a reference set of class-labeled blood-based mass spectral feature-table data with Early/Late labels, generating hierarchical multi-level outputs using kNN mini-classifiers filtered by a threshold number of proposed classifications and combined into master classifiers, performing first-stage Early/Late stratification followed by second-stage stratification for platinum-refractory/platinum-resistant likelihood, and optionally constructing multi-stage and subgroup-partitioned hierarchies using clinical data to generate and combine classifiers C1…CN as a hierarchical multi-level classifier.
Stated Advantages
Not explicitly described in patent.
Documented Applications
Not explicitly described in patent.
Interested in licensing this patent?