Researchers have developed smart models that could help diagnose amyotrophic lateral sclerosis (ALS) earlier using biomarkers in the blood. The models have the potential to shorten the time to diagnosis of ALS, assess the severity of the disease and identify possible therapeutic targets.
Rapid diagnosis of neurodegenerative diseases remains a major challenge, especially when early symptoms are nonspecific and overlap with other conditions. New approaches using blood tests and advanced computer methods could help doctors recognize these diseases earlier and better predict their progression.
Michigan Medicine researchers have identified, with the help of machine learning models, a possible method for early diagnosis of amyotrophic lateral sclerosis based on a blood sample, according to a study recently published in the journal Nature Communications.
The models analyze the blood for biomarkers through gene expression, using RNA sequencing, in order to detect the presence of ALS. The same approach has the potential to predict the severity of the disease and the length of survival of patients with this neurodegenerative condition.
The authors point out that the results offer the opportunity to diagnose ALS earlier, which could allow patients to access treatments and clinical trials for which they would otherwise be ineligible due to the advanced stage of the disease.
Currently, ALS patients usually survive between two and four years after diagnosis. However, the disease is difficult to recognize, especially in the initial stages, because many of the early symptoms can be confused with other more common neurological problems. For this reason, obtaining an official diagnosis can take over a year, and patients can be subjected to unnecessary investigations and procedures.
Instead of looking for a single ALS-specific biomarker, the Michigan Medicine team developed a genetic classifier capable of identifying multiple biomarkers of the disease, with the goal of speeding up diagnosis. The tool used, called a set of genetic expression biomarkers, is currently used in oncology, for example in the diagnosis of breast cancer and the classification of tumor subtypes.
The researchers identified more than 2,500 genes that show different expression in ALS patients compared to the control group, many of which are associated with the immune system. The data was fed into an XGBoost machine learning model, trained to estimate the presence of SLA. After narrowing down the biomarker combination to sets of between 27 and 46 genes, the model predicted ALS with up to 91% accuracy.
According to the authors, the model was tested both on biological samples collected from the patients included in the study and on data from other research, and obtained better results than previous methods for identifying a biomarker signature for ALS.
Later, the team developed two more sets of biomarkers, using other machine learning models, to predict survival in ALS. This time, in addition to gene expression levels, clinical information was also included, which allowed better differentiation between cases with short, intermediate and long survival.
Previous research has linked protein levels called neurofilament light chains (NfL), an indicator of neuronal damage, with disease progression, but the study authors note that NfL is also elevated in other neurodegenerative diseases, such as Alzheimer’s, Parkinson’s and multiple sclerosis, which limits its specificity.
The analysis also highlighted the existence of “specific genes” in the blood of people with ALS, which share characteristics with motor neurons in the spinal cord, those mainly affected by the disease. Using these characteristic genes, the researchers identified eight drugs with a potential therapeutic role in ALS.
Some of these drugs, such as the antipsychotic trifluoperazine and the BTK inhibitor ibrutinib, have already been investigated in previous studies as possible therapeutic options for ALS.
Currently, there is no clinically developed biomarker for ALS prognosis. The authors emphasize that further research is needed to confirm the usefulness of the model in increasing diagnostic accuracy and shortening the time to diagnosis, and to validate the results regarding potential drug targets before they can be applied in clinical practice.