It's important to note that models that use genome wide association analysis have demonstrated extremely high predictive value across large cohorts sharing geography but are very poor when applied on a geographically distinct population. This suggests that although autism has a strong association with genetics, neurophysiology unique to autism develops in the context of highly complex genetic associations that are likely subject genetic drift across population and time.

When we have a a few genes of interest that are important in screening for a rare disease we accept that novel variants will continue to be identified throughout the years as more people are screened. Autism as a prevalence of 1-3%. I don't remember the exact number but I think something like 30% of autism diagnoses are believed to be secondary to fairly severe but distinct genetic syndromes. So when we talk about a subgroup of autism without clear etiology we are looking at a fraction of 70% of 3% of the population. We're approaching rare disease territory when we talk about subgroups within autism. A rare disease with a highly complex genetic association across the genome that is subject to genetic drift is not a good candidate for genetic screening.

All that being said, studies like this may provide valuable insight into what microbiology is being influenced, even if we can't reliably predict which variants are responsible. Id love to see future investigations relate genetics to biomarkers instead of behavioral tests in autism.