A multi-omics, multi-algorithm approach will unleash a new wave of understanding around Alzheimer’s disease and diagnosis, that in time, will make precision diagnostics practical, speakers predicted at the Alzheimer’s Association International Conference 2020.
A multi-omics, multi-algorithm approach will unleash a new wave of understanding around Alzheimer’s disease and diagnosis, that in time, will make precision diagnostics practical, speakers predicted at the Alzheimer’s Association International Conference 2020 virtual event, held July 27-31. At least a part of that understanding will be based on the biomarkers identified during this meeting.
For analytics, a multi-tool approach “is moving the emphasis from determining which algorithm is better to how to put the tools together (to get the most accurate information possible), Christos Davatzilos, Ph.D., University of Pennsylvania School of Medicine, said during a live chat. That enables researchers to be less focused on the engineering aspects and thus to work as translational scientists as more is learned about the brain and neurological diseases.
“There are different criteria for performance than for triaging individuals based on biomarkers,” Duygu Tosun-Turgut, Ph.D., University of California – San Francisco, said. “We need to be very mindful of the algorithms and how they are meant to be used – for clinical practice of translational trials.”
Therefore, as Davatzilos said, “It’s difficult to say any one algorithm is the best, but you can say this is best under certain conditions or for certain problems. Ultimately, AI won’t be very different from biostatistics now, in that there are many options and many criteria researchers need to consider when choosing their tools.”
So far, vetting various algorithms and approaches has received a good deal of attention.
“We need such a diversity of means of discovery. With a multi-omics approach, researchers have independent sources of validation they can use to verify (their results) using a separate approach,” said Paul M. Thompson, Ph.D., University of Southern California.
Initially, Thompson recalled, “We learned a lot from the psychiatric researchers. They advised against using physical self-maintenance (PSM) scores,” as criteria during the first round of analysis because of the high percentage of false positives those scored could generate. Now that more computing power is available and machine learning can be applied to the research, that advice becomes less valid because researchers have the means of sorting through the data.
“There will be a new wave of discovery when people accept machine learning,” Dr. Thompson added.
Machine learning opens the door to increasingly predictive approaches. For example, Andrew Saykin, Psy.D., director of the Indiana Alzheimer Disease Center, pointed out, “We’re being so stringent that we’re filtering out things that may be important in our concern to eliminate noise.”
As an example, Thompson mentioned an MRI study of 11,000 young to middle-aged people.
“We didn’t find apolipoprotein E (APOE),” the strongest known genetic risk factor for developing Alzheimer’s disease. That study shows that subjects’ ages matter, because APOE becomes more prevalent in people after age 60. It also suggests that, as Thompson said, “If you build in interactions or other conditions, such as druggable characteristics, there may be better filters than the statistical sledgehammer” that is being used today.
One possibility is a better understanding of the effects of various risk factors on cognitive trajectories in early- and late-onset Alzheimer’s disease. A team from South Korea’s Sungkyunkwan University, the Samsung Medical Center, Samsung Alzheimer Research Medical Center, and in the US, from Harvard Medical School and Indiana University found that conventional risk factors accelerated cognitive decline in late-onset Alzheimer’s but slowed the decline among early-onset patients.
In studying the effects of APOE4, they found that outcomes varied based upon whether the patients carried that protein. Specifically, early-onset patients who did not carry APEO4 had a more rapid cognitive decline, “but not those with late-onset Alzheimer’s disease.” Likewise, higher levels of education accelerated cognitive decline in both early- and late-onset patients, but the most pronounced effects were among those with early-onset Alzheimer’s disease.
This research team also found that hypertension and obesity slowed cognitive decline in early-onset Alzheimer’s patients, but not in late-onset patients. Not surprisingly, the team noted that additional research is needed to determine the genetic and environmental factors behind declines in early-onset patients who lack any known risk factors.
In other meeting highlights, a poster presented Tuesday by Xiaohui Yao, Ph.D., and Li Shen, PhD., both of the University of Pennsylvania, predicted tissue-specific gene targets for Alzheimer’s disease. The researchers combined genome-wide association studies and data from expression quantitative trait loci (eQTL) to identify several genes in various tissues that have multiple associations (pleiotrophy) between their expression levels and the diagnosis of Alzheimer’s disease.
Fourteen genes and 13 brain tissues showed 64 significant associations. The gene CTB-171A8.1 exhibited the highest level of associations, with log p values typically beyond 10m and, for specific tissues, often above 12.5. The highest levels were associated with the cortex, cerebellum, and caudate basal ganglia, although activity was exhibited in all 13 brain tissues. The next most active gene was CEACAM19, which was involved in 10 of the tissues. Yao and Shen plan to continue investigating these genes as targets for functional validation.
