Methods of treating a subject suffering from rheumatoid arthritis based in part on a trained machine learning classifier
Inventors
Ghiassian, Susan • Mellors, Theodore R. • Santolini, Marc • Ameli, Asher • Schoenbrunner, Nancy • Akmaev, Viatcheslav R. • Johnson, Keith J.
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-11456056-B2
Publication Date
2022-09-27
Expiration Date
Abstract
Presented herein are systems and methods for developing classifiers useful for predicting response to particular treatments. For example, in some embodiments, the present disclosure provides a method of treating subjects suffering from an autoimmune disorder, the method comprising a step of: administering an anti-TNF therapy to subjects who have been determined to be responsive via a classifier established to distinguish between responsive and non-responsive prior subjects in a cohort who have received the anti-TNF therapy.
Core Innovation
The invention provides systems and methods to predict rheumatoid arthritis responsiveness to anti-TNF therapy using a trained machine learning classifier. Responsiveness versus non-responsiveness is predicted based at least in part on analyzing blood RNA expression levels of a set of genes including ARPC1A, ATAT1, CD27, MORN2, SNX8, SSNA1, and UBL7-AS1.
The trained machine learning classifier distinguishes between responsive and non-responsive subjects who have received the anti-TNF therapy, with performance constraints such as an area under the curve of at least about 70% and, in other independent claims, an accuracy of at least about 90%. The prediction may further be based on the presence of one or more SNPs in an expressed gene sequence and/or at least one clinical characteristic of the subject.
The disclosure also supports treating rheumatoid arthritis by administering an alternative to anti-TNF therapy to a subject predicted to be non-responsive, and administering the anti-TNF therapy to a subject predicted to be responsive. The classifier includes either a neural network or a random forest, and may be trained using rheumatoid arthritis subjects categorized as responsive versus non-responsive to anti-TNF therapy.
Claims Coverage
The independent claims are directed to treating rheumatoid arthritis subjects using anti-TNF therapy versus an alternative therapy based on predictions from a trained machine learning classifier. Across the independent claims, there are four inventive features.
Anti-TNF administration guided by ML responsiveness prediction with AUC constraint
Administering anti-TNF therapy to a subject with rheumatoid arthritis that has been predicted to be responsive based at least in part on a trained machine learning classifier that distinguishes between responsive and non-responsive subjects, with an area under the curve of at least about 70%.
Alternative therapy for ML-predicted non-response based on named gene set
Administering an alternative to anti-TNF therapy to a subject predicted to be non-responsive based at least in part on a trained machine learning classifier distinguishing responsive versus non-responsive subjects based at least in part on analyzing an expression level of a set of genes comprising ARPC1A, ATAT1, CD27, MORN2, SNX8, SSNA1, and UBL7-AS1.
Treatment decision using received expression level, ML prediction, and alternative vs anti-TNF administration
Receiving an expression level in the subject of a set of genes comprising ARPC1A, ATAT1, CD27, MORN2, SNX8, SSNA1, and UBL7-AS1; predicting whether the subject is responsive or non-responsive to an anti-TNF therapy based at least in part on a trained machine learning classifier distinguishing responsive and non-responsive subjects based at least in part on analyzing the expression level in the subject; and administering an alternative to anti-TNF therapy if determined non-responsive, or administering anti-TNF therapy if determined responsive.
Anti-TNF administration guided by ML responsiveness prediction with accuracy constraint
Administering anti-TNF therapy to a subject with rheumatoid arthritis that has been predicted to be responsive based at least in part on a trained machine learning classifier that distinguishes between responsive and non-responsive subjects who have received anti-TNF therapy at an accuracy of at least about 90%.
Across the independent claims, treatment hinges on predictions from a trained machine learning classifier distinguishing responsive versus non-responsive subjects to anti-TNF therapy. The core inventive framework combines gene-expression analysis using a named set of genes, with optional additional inputs such as SNP presence and/or clinical characteristics, classifier implementation as a neural network or a random forest, and performance constraints including AUC at least about 70% and/or accuracy at least about 90%.
Stated Advantages
Predictive performance targets for distinguishing responsive versus non-responsive subjects, including area under the curve of at least about 70% and/or accuracy of at least about 90%.
Capability to guide administering anti-TNF therapy for subjects predicted to be responsive and administering an alternative to anti-TNF therapy for subjects predicted to be non-responsive.
Classifier prediction may achieve negative predictive value constraints of at least about 85% and true negative rate constraints of at least about 60%.
Documented Applications
Treating a subject suffering from rheumatoid arthritis by administering anti-TNF therapy when predicted to be responsive based on a trained machine learning classifier distinguishing responsive versus non-responsive subjects.
Treating a subject suffering from rheumatoid arthritis by administering an alternative to anti-TNF therapy when predicted to be non-responsive based on a trained machine learning classifier.
Interested in licensing this patent?