Machine Learning Models Diagnose ALS Earlier Through Blood Biomarkers

Early symptoms often overlap with other neurological conditions, leading to diagnostic delays that can exceed a year and limit access to treatments and clinical trials. Patients typically survive only two to four years after diagnosis, making early and accurate detection critical. Now, a new blood-based approach suggests that ALS may be identified earlier and disease progression better predicted using gene expression patterns analyzed with machine learning.

Researchers at Michigan Medicine (Ann Arbor, MI, USA) have developed machine learning models that analyze RNA sequencing data from blood samples to identify gene expression signatures associated with ALS. Rather than relying on a single biomarker, the team created a multi-gene expression biomarker panel similar to those used in oncology. More than 2,500 genes showing different expression patterns in ALS patients compared with controls were identified, many linked to immune system activity. These data were used to train machine learning algorithms to detect ALS-related signals.

The primary diagnostic model used XGBoost, a gradient-boosting machine learning method. Gene panels were refined to include between 27 and 46 genes, balancing accuracy with practicality. Additional models were later developed to predict survival by combining gene expression data with clinical information, enabling stratification into shorter, intermediate, and longer survival groups. This approach differs from existing markers such as neurofilament light chain, which lacks disease specificity and is elevated in multiple neurodegenerative disorders.

When validated using internal samples and external datasets, the diagnostic model predicted ALS with accuracy rates of up to 91%. The prognostic models successfully differentiated survival trajectories, addressing a major unmet need in ALS care. Further analysis revealed a set of “core genes” in blood that shared features with affected motor neurons in the spinal cord. Using these genes, researchers identified eight existing drugs with potential therapeutic relevance to ALS, including compounds previously linked to ALS-related pathways.

The findings, published in Nature Communications, suggest that blood-based gene expression profiling could reduce diagnostic delays and provide early prognostic insights, allowing patients earlier access to tailored care and clinical trials. The ability to predict survival may also help guide treatment planning and patient counseling. Researchers emphasize that additional studies are needed to validate the models in larger and more diverse populations. Future work will also explore the therapeutic potential of the identified drug candidates and further refine the models for clinical implementation.

“Our findings present an incredible opportunity to potentially diagnose ALS earlier, which opens up doors to treatments and clinical trials for which people otherwise may not be eligible due to advanced disease,” said co-senior author Eva L. Feldman, M.D., Ph.D.

Related Links:
Michigan Medicine