A new study combining a multi-omics approach with machine learning has identified changes in plasma proteins that could allow early detection of amyotrophic lateral sclerosis (ALS) long before clinical symptoms occurred. Works published in Natural Medicine August suggests that plasma biomarkers can be used to identify ALS in potentially asymptomatic individuals 10 years before the onset of symptoms.
ALS remains difficult to diagnose early
ALS, sometimes called Lou Gehrig's disease, is a rare neurodegenerative condition that gradually affects neurons in the brain and spinal cord. Over time, patients lose spontaneous muscle control, affecting speech, movement and breathing. Currently, there are no conclusive diagnostic tests, and diagnosis often occurs later in the disease progression. On average, patients wait 6-18 months for diagnosis, most survive only 2-4 years after onset of symptoms.
Multiomics and machine learning identify early protein changes
The research team based at the US National Institutes of Health analyzed almost 3,000 plasma proteins in patients with ALS. From this group, we found that 33 proteins were significantly different compared to controls. In particular, changes in muscle, nerve, and energy metabolic proteins were observed up to ten years before symptoms appeared.
Between the patients carrying C9ORF72 Gene expansion, a known genetic risk factor for European ancestors, has been found to be consistently elevated. These included eIF2S2, HPCAL1, JPT2, MTIF3, PDAP1, and SMAD3. This study proposes that these proteins may serve as indicators of early disease progression, especially in genetically predisposed individuals.
The Random Forest model shows high diagnostic accuracy
To interpret the data, researchers evaluated 10 machine learning models and found that the random forest method is best suited to classify cases of ALS. This model incorporated 17 protein biomarkers as well as demographic and technical variables such as age, gender, and type of blood collection tube used.
Initial testing of the model yielded a high area under the curve (AUC) score of 96.2%, resulting in a balanced accuracy of 89.3%. These metrics demonstrate a strong ability to distinguish ALS from non-ALS samples. Further validation in an external cohort of over 23,000 individuals confirmed its performance with an average accuracy of >98%.
Early intervention may be possible
The analysis also showed that the model could estimate the onset time of symptoms based on plasma protein signatures. These biological changes appear to reflect the compensation process that occurs before the outward signs of disease are present.
However, the study authors pointed out that their current proteomics platform may not capture changes in all proteins associated with ALS. The ongoing work aims to integrate a wider range of proteomic technologies and longitudinal data to improve the predictive power of the model.
Genome sequencing can support early risk detection
This finding is consistent with broader efforts to integrate genomic sequencing into routine health care. For example, the UK's neonatal genome program, supported by £650 million government funding, plans to provide whole-genome sequencing (WGS) to all neonates. This approach helps identify individuals who have genetic mutations. C9ORF72the risk of ALS and other conditions is high.
The WGS platform, like BGI Genomics, provides a detailed view of the genome, supporting early diagnosis and individualized health plans. Integration of protein biomarker data and genomic information could ultimately provide a framework for monitoring disease risk over lifespan.
reference: Chia R, Moaddel R, Kwan JY, and other plasma proteomics-based candidate biomarker panels predict amyotrophic lateral sclerosis. Nut Med. 2025. doi:10.1038/s41591-025-03890-6
This article has been republished from material. Note: Materials may have been edited for length and content. For more information, please contact the source quoted. You can access the press release publishing policy here.
This content contains text generated with AI support. You can find the AI policy for your technology network here.
