Knowledge-guided machine learning for rare disease diagnosis using electronic health records
Active Project Grant Awarded Last action
- Award ID
- F31HL184744
- Recipient
-
President And Fellows Of Harvard College UEI UNVDZNFA8R29
- Awarding agency
-
National Heart Lung and Blood Institute
National Institutes of Health, Department of Health and Human Services
- Obligated
- $44,106
Award facts
- Assistance listing
- 93.837 Cardiovascular Diseases Research
- Completion date 0% complete
- Place of performance
- Boston, MA 02115, USA
Award details come from USAspending.gov.
Description
- Abstract
- Abstract rare diseases, broadly defined as conditions affecting fewer than 1 in 2,000 people, collectively impact over 300 million individuals worldwide. Despite their broad reach, timely and accurate diagnosis remains a significant challenge due to the low prevalence of individual conditions and limited familiarity among healthcare providers. As a result, many patients experience a years-long "diagnostic odyssey" marked by inconclusive testing, repeated misdiagnoses, and missed opportunities for early intervention. Rare pulmonary diseases - including pulmonary hypertension, bronchiectasis, and interstitial lung disease - exemplify these challenges. Their early manifestations often resemble those of more common respiratory disorders such as asthma, lead- ING to frequent misdiagnosis and delayed initiation of appropriate care. Strengthening diagnostic pathways for rare pulmonary diseases requires scalable, data-driven approaches capable of detecting subtle phenotypic patterns often overlooked in routine clinical care. Electronic health records (EHRS) contain rich, longitudinal data that can support data-driven phenotype detection and discovery. However, effective computational phenotyping from EHRS is hindered by inconsistent coding practices, missing data, and a scarcity of reliable diagnostic labels to guide inference. While recent advances in machine learning have improved phenotyping for common conditions, their application to rare disease research remains constrained by limited sample availability, variable data quality, and poor model transferability across health systems. This proposal develops two complementary knowledge-guided machine learning frameworks to address these limitations by embedding sparse, heterogeneous patient data within a broader biomedical context. The first is a weakly supervised transformer framework that integrates a limited number of expert-validated labels with probabilistic EHR-derived labels and pre-trained biomedical embeddings to improve phenotype prediction and patient stratification. The second is a knowledge-guided graph framework that regularizes EHR-derived CO-occurrence networks using structured biomedical knowledge - including both disease-specific and general- purpose knowledge graphs - to enable robust and interpretable rare disease inference. We will evaluate the accuracy and generalizability of these methods in rare pulmonary diseases using EHR data from boston children's hospital and mass general brigham, focusing on early identification, subgroup discovery, and disease progression modeling. All code will be released as open-source software to promote reproducibility and broader adoption across rare disease applications. By embedding noisy, real-world clinical data within a knowledge-guided framework, this work aims to shorten diagnostic delays and enable more timely, personalized care for patients with rare pulmonary diseases.
Quick answers
Who received award F31HL184744?
F31HL184744 was awarded to President And Fellows Of Harvard College by National Heart Lung and Blood Institute (National Institutes of Health) on January 31, 2027.
How much is award F31HL184744 worth?
$44,106 has been obligated on F31HL184744.
On GovTribe
See this award's full record
- Every transaction and subaward
- The recipient's other awards and teaming partners
- Similar grant awards
- Ask GovTribe AI about this award