Systems and temporal alignment methods for evaluation of gestational age and time to delivery
Inventors
Liang, Liang • Samyak, Rajanala • Tibshirani, Robert • Hastie, Trevor • Snyder, Michael P.
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-12426782-B2
Publication Date
2025-09-30
Expiration Date
Abstract
Methods to compute gestational age and time to delivery utilizing temporal alignment methods and applications thereof are described. Generally, systems utilize analyte measurements collected at one or more time points to determine a gestational age and time to delivery, which can be used as a basis to perform interventions and treat individuals. Computational models trained utilizing temporal alignment of analyte measurements can be used to determine gestational age and time to delivery.
Core Innovation
The invention provides methods for determining gestational age or a time to delivery of a pregnant individual by collecting biological samples from the pregnant individual at two or more time points and measuring one or more analytes from each time-point sample. The method yields analyte measurements for each time point and pairs each set of analyte measurements with its time point of collection to produce a data feature used as input into a computational model.
The computational model is trained utilizing temporally aligned data features provided by a cohort of pregnant individuals. Each training data feature comprises a set of analyte measurements and its time point of collection, and each pregnant individual of the cohort provides two or more data features. The model predicts gestational age or time to delivery using temporally aligned predictions of the pregnant individual and temporally aligned predictions of each pregnant individual within the training cohort.
Each temporally aligned prediction is an error-corrected prediction corrected by temporal differences between the two or more data features. An alternative modeling approach uses a principal-curve computational model that fits a one-dimensional feature-space curve as a function of time from the collected data features, and prediction deviation is optimized based on where the data features are mapped along the one-dimensional curve.
The disclosure also ties the determined gestational progress to downstream diagnostics and interventions/treatments, and describes kits for detecting biomarker panels to monitor pregnancy, including risk assessment for preterm birth and spontaneous abortion.
Claims Coverage
The provided independent claim describes a temporally aligned, error-corrected computational-model approach using (analyte measurements, timepoint) features to predict gestational age or time to delivery. It includes inventive elements for the training procedure, the temporal alignment of predictions, and the use of constrained error correction by temporal differences, with optional refinements specifying principal-curve modeling and selected metabolite features.
Temporally paired analyte feature input
Collecting a biological sample of a pregnant individual from each time point of two or more time points; measuring one or more analytes of each biological sample to yield a set of analyte measurements for each time point; and determining a gestational age or a time to delivery via a computational model using paired data features where each set of analyte measurements and its time point of collection are paired to yield a data feature utilized as input into the computational model.
Cohort-trained computational model using temporally aligned data features
Training the computational model utilizing temporally aligned data features provided by a cohort of pregnant individuals, where each data feature for training comprises a set of analyte measurements and its time point of collection, and each pregnant individual of the cohort has provided two or more data features.
Error-corrected temporally aligned prediction by temporal differences
Predicting the gestational age or the time to delivery using temporally aligned predictions of the pregnant individual and temporally aligned predictions of each pregnant individual within the training cohort, wherein each temporally aligned prediction is an error-corrected prediction corrected by temporal differences between the two or more data features.
Principal-curve one-dimensional time curve mapping and deviation optimization
Using a principal curves computational model to produce a one-dimensional time-based curve from collected data features; generating predictions for each pregnant individual from the mapped features; and optimizing prediction deviation based on where the features map along that curve.
Selected metabolite model feature
Where the model feature is a measurement of at least one metabolite selected from THDOC, estriol-16-glucuronide, and progesterone.
Across the provided independent claim and cited refinements, the coverage centers on using paired (analyte measurements, timepoint) features, training on a cohort with temporally aligned data features, and predicting via temporally aligned error-corrected predictions corrected by temporal differences. Additional refinements provided in the partial content include principal-curve modeling and selecting specific metabolite measurements (THDOC, estriol-16-glucuronide, and progesterone) as model features.
Stated Advantages
Improved prediction accuracy is emphasized through temporally aligning multiple subject-specific predictions using sampling-time differences and using temporally aligned, error-corrected predictions.
The principal-curves approach provides a structured one-dimensional time-based modeling of collected data features and maps aligned predictions to the curve.
Documented Applications
Downstream diagnostics and interventions/treatments based on the determined gestational progress.
Monitoring pregnancy using kits for detecting biomarker panels, including risk assessment for preterm birth and spontaneous abortion.
Interested in licensing this patent?