Multi-omics machine learning can detect ALS 10 years before symptom appears

(Press-News.org) A new study applying multi-omics techniques and machine learning identified 33 plasma proteins that differ significantly in patients with amyotrophic lateral sclerosis (ALS). The findings suggest ALS could be detected up to 10 years before symptoms appear, offering promise for reliable diagnostic biomarkers.

The integration of advanced high-throughput sequencing technologies, such as whole genome sequencing (WGS) for the early detection of rare diseases, such as ALS, could give clinicians and patients a critical window for medical intervention, ultimately improving outcomes.

Trapped in a Paralyzed Body

ALS, formerly known as Lou Gehrig's disease, is a rare neurodegenerative condition. It attacks nerve cells in the brain and spinal cord, gradually robbing people of their ability to walk, talk, and eventually breathe.

Public awareness of ALS grew rapidly in 2014 with the Ice Bucket Challenge. It was a social media campaign that became global phenomenon where celebrities, politicians, influencers, and everyday people pouring buckets of ice water over themselves to raise attention for the disease. The ASL Association praised the campaign for “dramatically accelerating the fight against ALS.”

Despite advances in neurophysiology and genetic testing, there is still no definitive diagnostic test. Patients often wait 6 to18 months for a diagnosis while the disease progresses rapidly. Most patients survive only 2 to 4 years after symptom onset.

Diagnosis Before Symptoms Appear

Researchers at the U.S. National Institutes of Health (NIH) analyzed nearly 3,000 plasma proteins and identified 33 that were significantly different in patients with ALS.

The study, published in Nature Medicine in August, points toward earlier and more accurate detection of the disease.

One of the most meaningful discoveries was that disease-related changes in skeletal muscle, nerves, and energy metabolism began up to 10 years before symptoms appeared. This opens the possibility for earlier diagnosis, earlier intervention, and potentially more effective treatments.

The researchers also studied ALS patients carrying a C9orf72 gene expansion, a common genetic factor for ALS in people of European ancestry. They found eight proteins significantly elevated in this group, including EIF2S2, HPCAL1, JPT2, MTIF3, PDAP1, and SMAD3. These proteins may serve as early biomarkers of disease progression.

Machine Learning Supports Risk Prediction

The researchers compared 10 different machine learning approaches and found that a random forest model, a type of algorithm that combines many decision trees, worked best for ALS detection. This model used 20 factors in total, including 17 plasma proteins as well as age, gender, and even the type of blood collection tube.

When tested, the model showed strong performance, with an area under the curve (AUC) of 96.2% and a balanced accuracy of 89.3%. These numbers mean the model was highly reliable in separating ALS cases from non-ALS cases.

The team then validated their model in a much larger group of more than 23,000 individuals. The model maintained extremely high accuracy of over 99%, and could clearly tell ALS apart from other neurological or muscle-related disorders. Across both external validation groups, the model reached an average accuracy of 98.3%, making it a potential tool for diagnosis.

Beyond diagnosis, the researchers asked whether the model could predict when ALS symptoms would begin. The results showed a clear link between the plasma protein-based “ALS risk score” and the eventual onset of symptoms. Remarkably, protein changes could be detected up to 10 years before the first signs of ALS, suggesting that the body undergoes compensatory changes long before patients notice anything wrong.

The team noted limitations as their protein analysis platform did not capture all possible changes. They plan to expand future studies with broader proteomics methods and longitudinal datasets to further refine their model.

WGS for All Newborns

The potential of early detection extends beyond ALS. In the UK, the Newborn Genomes Programme, backed by £650 million in government funding, will offer whole genome sequencing (WGS) to every newborn. The program aims to create a digital health record for lifelong monitoring.

WGS is also critical for detecting rare diseases and childhood cancers caused by genetic mutations. Services like BGI Genomics' human WGS provide a comprehensive, base-by-base view of the genome, delivering the data needed for early risk detection and personalized care.

For newborns carrying genetic mutations such as C9orf72, WGS could potentially warn of ALS risk decades before symptoms appear. This would allow clinicians to plan interventions that might slow disease progression, or even prevent it when new target therapeutics are available.

This study marks a significant breakthrough in ALS research. Plasma proteins may serve as early biomarkers, and machine learning models could transform diagnosis from reactive to predictive. Combined with advances in genome sequencing, rare conditions can be detected earlier and more effective interventions can be applied, leading to improved patient outcomes.

 

About BGI Genomics WGS

BGI Genomics' Human whole genome sequencing (WGS) service detects the complete genome sequence at one time and provides a high-resolution, base-by-base view of the genome. This enables researchers to see both large and small variants and identify potential causative variants for further follow-on gene expression or regulation mechanism studies. BGI Genomics' Human Whole Genome Sequencing services are typically executed with proprietary DNBSEQ™ sequencing technology platforms, for great sequencing data at some of the lowest costs in the industry. DNBSEQ™ can offer advantages in terms of lower amplification error rates and much lower duplication rates in WGS/WES applications.

About BGI Genomics

BGI Genomics, headquartered in Shenzhen, China, is the world's leading integrated solutions provider of precision medicine. Our services cover more than 100 countries and regions, involving more than 2,300 medical institutions. In July 2017, as a subsidiary of BGI Group, BGI Genomics (300676.SZ) was officially listed on the Shenzhen Stock Exchange.

END