Study results are preliminary, but they offer proof of concept that the method can link certain behavior, or phenotype, to a specific genetic structure or genotype. The signatures of shared behavior may indicate shared gene pathways that lead to behaviors, which in turn could hint at the cause of autism.
“The power of the machine learning of the vector support system is that you can find hidden patterns, ie patterns that were not detected by statistical analysis without conventional supervision” says Hilgo Bruining, assistant professor of child and adolescent psychiatry at the University of Utrecht, who also led the study.
The researchers plan to sift through large sets of behavioral and genetic data of individuals with idiopathic autism. If the algorithm can identify new behavior signatures inside these sets of data, it may be able to divide into subgroups of autism and concentrate on the genome areas responsible with the disorder subtypes.
However, some experts call for caution with this line of reasoning.
Valsamma Eapen, professor of child and adolescent psychiatry at the University of New South Wales in Australia, who was not involved in the study, says: “As should have been expected, the phenotypes of autism could be spaced around a diverse set of genetic lesions, suggesting that these results mean different types of autism are an exaggeration”.
For their study, the researchers analyzed behavioral data in the medical records of 322 people at the University of Utrecht Medical Center in the Netherlands and the Institute of Psychiatry at Kings College London. The subgroups diagnosed with each of the six autism-related syndromes, comprised between 21 and 90 people.
In particular, researchers worked with data from 37 behaviors, including verbal rituals, imaginative and unusual occupations in their system.
On the task of distinguishing between the six syndromes, the algorithm correctly identified a behavioral syndrome signature 63% of the time.
“Sixty-three percent is a bit low, but it’s not intended to be an absolute,” says Wall.
The algorithm was more accurate when two syndromes were compared simultaneously. For example, it could distinguish the 22q11.2 deletion syndrome from the Chromosome 15 isodicentric with 97% accuracy, its best performance. It has been worse when it comes to distinguishing the syndrome from five other disorders.
To test the algorithm in cases of autism, researchers look at behavioral data of 1,261 persons selected in the Autism Genetic Research Exchange (AGRE). However, this data does not include information on whether people are diagnosed with one of the six syndromes, so the algorithm can only generate a probability that each identification is correct.
Within the first sample of 375 people from AGRE, the algorithm classified 255 people with behavior signing of the tuberous sclerosis complex with a 61 percent chance that this analysis is correct. The figures for the other syndromes were lower, with several of them in the range of 40 percent and with the Down syndrome, again the lowest, at 25 percent probability of accuracy.
The researchers then repeated the test with two sets of 443 people each, yielding similar results.
In general, they say, the pattern of social difficulties is the most useful to help identify syndromes. This suggests that the social impairment, such as extreme social avoidance seen in people with fragile X syndrome may be related to certain genetic risk factors, the researchers say.
Eapen says he expects to see studies in which the cause of the disorders and the underlying mechanisms are better outlined.
“In these deletion syndromes, no one is sure that the genes are linked to autism and how it affects the size of the demonstrations and duplications of genes or phenotypes,” said Eapen. “This study sets the stage for a closer look at diagnosis in cohort selection based on genetic damage.”