UCLA researchers in the Department of Computational Medicine have developed a novel machine learning algorithm to generate an accurate molecular score for determining asthma drug response.
BACKGROUND: Asthma is a globally prevalent, high-burden disease marked by substantial clinical heterogeneity and variable treatment response. This heterogeneity arises from interindividual differences in disease mechanisms, shaped by genetic predisposition, environmental exposures, and their interactions. Defining biologically and clinically meaningful subtypes is central to precision medicine efforts in asthma. Molecularly defined subtypes—known as endotypes—have been associated with disease severity and clinical trajectory. Although endotype-based classification has advanced asthma care, treatment response remains inconsistent, and disease control suboptimal for many patients.
Among these endotypes, type 2 (T2)-high and T2-low asthma are distinguished by underlying immune mechanisms and inflammatory profiles, each associated with biomarkers that guide therapy. T2-high asthma is primarily driven by type 2 helper T (Th2) cells and group 2 innate lymphoid cells (ILC2s), which secrete interleukins (IL)-4, IL-5, and IL-13. These cytokines orchestrate allergic and eosinophilic inflammation, manifesting clinically as allergic sensitization and elevated blood eosinophil counts (BEC).
Most approved biologic therapies for moderate to severe asthma target the T2 inflammatory pathway. However, even when patients meet biomarker thresholds for T2 inflammation, responses to these costly targeted biologics—often exceeding $40,000 annually—are frequently incomplete. In a large clinical study by Hansen et al. (2024), fewer than 25% of adults with severe asthma achieved clinical remission despite biologic therapy. Similarly, Bacharier et al. (2021) demonstrated that among children with elevated BEC and IgE, treatment with dupilumab resulted in a 50% reduction in disease exacerbations compared to placebo. These findings highlight the limitations of current biomarkers for predicting treatment response and underscore the need for more precise stratification tools to uncover alternative disease mechanisms and inform novel therapeutic approaches.
INNOVATION: Researchers at UCLA led by Dr. Elior Rahmani have developed a novel machine learning algorithm that generates molecular diagnostic scores, which can be used to accurately determine response to asthma therapies. Two score models were defined: a model based on DNA methylation markers and a second model based on gene expression markers. These models can be combined or used separately.
The approach leverages uniquely large training and validation datasets of 1500+ ethnically and ancestrally diverse pediatric asthma patients using DNA methylation (DNAm) and gene expression profiles. Additional validation data includes adult asthma patients from a randomized clinical trial.
Unlike current clinical assessments of T2 asthma patients, which rely heavily on peripheral biomarkers such as BEC and IgE levels, this algorithm incorporates epigenetically encoded markers and/or gene expression markers to significantly enhance predictive accuracy. Consequently, this model generates scores that outperform traditional diagnostic standards and enables effective patient stratification by clinically meaningful subgroups of asthma patients. This enhances the ability to predict the response to common treatments like Albuterol and advanced biologics like Omalizumab, thereby reducing both the clinical burden of trial-and-error prescribing and the financial cost of ineffective treatments due to mischaracterized diagnoses.
POTENTIAL APPLICATIONS:
- Predicting drug response to anti-T2 therapies, such as omalizumab, dupilumab, and mepolizumab.
- Personalized biologic treatment strategies for T2-high asthma patients
- Accurate classification of asthma patient subgroups
ADVANTAGES:
- Cost-effective stratification of T2-high asthma patients prior to costly—often exceeding $40,000 annually—biologic therapies
- Increased racial and ethnic diversity of genomic training data for accurate and robust patient subgrouping
- Patient subgrouping was validated on both pediatric and adult patients
- Reduced healthcare costs from the use of ineffective asthma treatments
DEVELOPMENT-TO-DATE: The inventors developed their model using a large discovery dataset and replicated their results on three test datasets, including validation data from a randomized clinical trial with gene expression collected from participants, which confirmed their gene expression markers as prognostic of response to omalizumab. They have also showed their DNA methylation score is predictive of bronchodilator response with high precision.
Related Papers (from the inventors only):
Gorla A et al. Epigenetic patient stratification via contrastive machine learning refines hallmark biomarkers in minoritized children with asthma. Res Sq (Preprint). Sep 13:rs.3.rs-5066762 (2024)
Gorla A et al. Phenotypic subtyping via contrastive learning. bioRxiv (Preprint). Jan 6:2023.01.05.522921 (2023)
KEYWORDS: Asthma, Type 2 asthma, Type 2 inflammation, biologic therapy, machine learning, epigenetics, DNA methylation, biomarker, eosinophil, omalizumab, dupilumab, mepolizumab, bronchodilator response (BDR), precision medicine, Albuterol