Study: Machine learning uncovers new genes associated with ALS
Genes concern mitochondrial function; metabolizing fat, iron; forming vesicles
A machine learning analysis of gene activity in spinal cord samples from people with amyotrophic lateral sclerosis (ALS) revealed new genes associated with the neurodegenerative disorder, a study reports.
The newly identified genes included those involved in the function of energy-producing mitochondria, lipid (fat) and iron metabolism, and the formation of vesicles, or tiny, membrane-bound sacs that move substances into and out of cells.
“These genes are particularly intriguing because they point to new areas of investigation in ALS research,” the researchers wrote in the study, “Machine learning identified novel players in lipid metabolism, endosomal trafficking, and iron metabolism of the ALS spinal cord,” which was published in Scientific Reports.
ALS is marked by the progressive deterioration of motor neurons, the nerve cells in the brain and spinal cord that control voluntary movement and allow a person to move, speak, breathe, and swallow. While the disease has been associated with mutations in certain genes, most people with sporadic ALS who have no family history of the disease have no clear genetic cause.
“Although genes causing familial cases have been identified, those of sporadic ALS, which occupies the majority of patients, are still elusive,” wrote researchers in Taiwan, who analyzed gene activity data of spinal cord samples from 199 deceased ALS patients and 41 people without neurological disease to further investigate the genetics underlying ALS.
ID’ing new genes linked to ALS
For their analyses, the researchers used machine learning, a type of artificial intelligence that uses algorithms to learn from data and identify and predict patterns. The data served as input to create so-called binary classifiers capable of distinguishing between ALS and control samples.
After applying four different machine learning algorithms, the overall accuracy of these binary classifiers in differentiating between the ALS and control samples was more than 80%. One algorithm called the generalized linear model (GLM) achieved 91.7% accuracy.
To validate the findings, the team applied the trained classifiers to unseen gene activity data from 222 spinal cord samples from ALS patients and controls. The samples had been taken from the lumbar, or lower, part of the spinal cord, while the samples from the training dataset had been taken from the cervical, or upper part.
Across all four algorithms, the classifiers maintained a high level of accuracy, above 77% in all cases, with GLM achieving 89.6% accuracy. The classifiers’ ability to identify control samples alone varied, ranging from 34% to 83%, depending on the algorithm.
The researchers suggested the genetic patterns in the cervical spinal cord of ALS patients appeared to also be present in the lumbar spinal cord, making it easier for the algorithms to detect ALS regardless of the spinal cord region. On the other hand, the lower accuracy in control samples suggests there may be slight differences in the genetics of controls between the cervical and lumbar regions.
All four algorithms collectively identified 114 genes representing distinguishing features between ALS and control spinal cord samples. Of these, 41 had been reported to be associated with ALS, demonstrating the “machine-learning approach was successful in pinpointing key genetic markers that have already been associated with the disease,” the scientists wrote.
Among the 73 newly identified ALS-linked genes, eight have been linked to other neurological diseases, including Parkinson’s and Alzheimer’s disease. Moreover, 21 genes not previously associated with ALS were noted to be involved in lipid transport, ion channel function, and the formation of vesicles, suggesting they play an unknown role in ALS development.
One algorithm predicted ALS via lower activity of the mt-ND2 gene, which is vital for the function of mitochondria, the cellular organelles that generate energy. Another algorithm found that lower activity of a gene involved in iron metabolism, PACC1, predicted ALS. Low activity of this gene “may promote neurodegeneration by mediating abnormal accumulation of iron in the spinal cord,” the researchers wrote.
“Binary classifiers [built] by machine learning on spinal cord … data successfully differentiate ALS and control samples,” the scientists wrote. “This study identified novel genes in mitochondrial respiration, lipid metabolism, [vesicle] trafficking, and iron metabolism, which may promote the progression of ALS.”
About the Author
