Using small data to target Alzheimer’s

NetraMark CEO on using AI to identify novel targets for treatment of Alzheimer’s disease.

Despite the billions spent on research and development into treatments for Alzheimer’s disease, it continues to be a growing cause of death among the elderly. Indeed, the Alzheimer’s Association reports that, while deaths from heart disease fell by 7.8% between 2000 and 2018, deaths from Alzheimer’s increased by a staggering 146% over the same period.

Longevity.Technology: With scientists realising that the biology of the disease is far more complex than first thought, there is an increasing focus on identifying the wide range of underlying mechanisms that may be contributing factors to Alzheimer’s. We spoke to Joseph Geraci, CEO of AI firm NetraMark, about his company’s recent work to identify those mechanisms from small patient datasets.

“Of course, it’d be great if we can come up with one drug that can help everybody with Alzheimer’s,” he says. “But the truth is that it’s a variety of disorders, which all look the same, and in the end there is the pathology that turns out to be this devastating dementia. But there’s a lot of ways to get there.”

Geraci points out that even someone in their mid-40s could be developing issues with their molecular circuitry that will lead to a type of dementia down the line. He imagines a future where these issues could be identified early in life, but recognises we’re a long way from that.

“The problem is that we would have to design a mega study to get there,” says Geraci. “But in the elderly, it’s easier because there’s lots of [small data] samples out there already. And we can use artificial intelligence to find out what’s going on.”

“… many methods of AI learn based on what you already know… you can also do discovery based on what you don’t know.”

While many methods of AI learn based on what you already know, Geraci points out that you can also do discovery based on what you don’t know.

“At NetraMark, we’ve developed a technique that’s really good at co-operating with you,” he explains. “So the machine says, ‘Okay, I know what you want me to learn, but you’re using me, because you want me to learn something you don’t know’. So in this interactive process, it hands you these hypotheses, and then you can literally use your mouse to evaluate these.”

By using NetraMark’s AI on sample data from a few hundred people with Alzheimer’s, Geraci’s team was able to identify a “menu” of six perspective classes that it hypothesises represent biological mechanisms that may act alone, or in combination to manifest an Alzheimer’s pathology. Genetic pathways identified include those associated with vasculogenesis, cellular signalling and differentiation, metabolic function, mitochondrial function, nitric oxide and metal ion metabolism.

“There could be more, but just from this sample we were able to see that there are these menu items, and from this buffet, we found that people’s pathologies were combinations of these in unique subtypes,” says Geraci. “And so what this means is that there are a lot of ways to get to dementia, which makes sense because the human biology is such a complex molecular machine.”

“…we’re trying to provide this landscape in neurodegeneration so that we can start attacking things very precisely.”

“And that’s why when you do a clinical trial with a single drug, you’re going to fail because 20% are responding in this way and another 15% that way, and so on. So what we’re doing now is cutting it up into subtypes, and trying to provide this landscape in neurodegeneration so that we can start attacking things very precisely.”

Geraci hopes that this work will ultimately lead to ways of early detection, blood tests for example, that will allow people to start making adjustments earlier in life to try to circumvent problems later in life, whether through diet, supplementation or drugs.

“Because we’re able to de-noise the data, we’re going to be able to say that you fall into this subtype, so this is your intervention,” he says. “The real vision is to continue down this path, and really dissect neurodegeneration into its components in such a way that we could actually recommend novel ways to drug the disease, either with repurposing or by partnering with a group that is particularly good with chemistry.”

“And we have this edge, because the technology we use is not based on standard methods, it’s able to learn from smaller data, there’s a lot of small data out there.”

NetraMark’s Alzheimer’s project is the subject of a recent preprint paper, and was conducted in collaboration with Rhoda Au, professor of neurobiology and neurology at Boston University. The paper is currently going through the peer-review process at a scientific journal.