Can AI cure Alzheimer’s disease?

Newsletter Signup - Under Article / In Page

"*" indicates required fields

A couple of weeks back, we took a look at how artificial intelligence (AI) has had a transformative impact on cancer research. Today, we take a look at how AI tools help tackle another formidable challenge of modern biotech, Alzheimer’s disease.

“Alzheimer’s is an incredibly complex disease with numerous factors at play. You have the buildup of amyloid plaques, tau tangles, and the loss of connections between nerve cells. If you put yourself in the shoes of a drug maker, how do you choose which one to target, and how do you determine the sequence of events in any given pathway?” said Nadia Harhen, general manager of AI simulation at SandboxAQ.

With traditional research methods alone proving insufficient, AI has emerged in Alzheimer’s research, tackling hurdles in drug discovery and diagnostics. AI’s ability to analyze massive datasets and model biological processes opens opportunities for early detection, understanding disease mechanisms, and even predicting drug efficacy.

The field already benefited from significant improvements, but can AI bring even more to Alzheimer’s research?

Diagnosing Alzheimer’s earlier and more efficiently with AI

“So far, the area of Alzheimer’s where AI has had the most impact is likely in diagnosis and detection. AI-assisted medical imaging allows us to detect the disease much earlier,” pointed out Harhen.

AI has significantly impacted Alzheimer’s diagnostics, particularly through assistance in medical imaging and biomarker analysis. Advanced machine learning models are now trained to detect early signs of Alzheimer’s in brain imaging scans, such as amyloid plaques, with impressive accuracy.

Indeed, a study by the University of California, San Fransisco (UCSF) showed that AI can analyze patient records to predict Alzheimer’s up to seven years before traditional clinical symptoms manifest, making it a game-changer for early detection and opening doors to earlier interventions.

Beyond imaging, AI holds vast potential in analyzing biomarkers – measurable indicators in blood, cerebrospinal fluid, and genetic data that could predict Alzheimer’s pre-symptomatically. “More and more people are working on predictive tools but if I had to identify an opportunity, I see very few working on biomarkers,” confirmed Harhen.

Similar to how AI contributed to oncology research, it is in diagnosis and prediction that these tools have transformed the field the fastest. But another evident area where AI has a lot to offer to Alzheimer’s research is in drug discovery.

Alzheimer’s drug discovery: Can AI change the paradigm?

“Beyond diagnosis, another transformative change is in drug discovery; we can now simulate how hypothetical drugs work at a molecular level and screen large sets of compounds much faster,” said Harhen.

But how does it work? Some of SandboxAQ’s work in the neurodegenerative space has been with Nobel Prize laureate, Dr. Stanley Prusiner, Harhen told Labiotech. She explained how the process went and how AI helped.

“Dr. Prusiner had several hypotheses about how Alzheimer’s progresses. What we did with computational AI was not only to replicate what they were seeing in the lab – which is often looking at proxies and combining different pieces of data to form a complete picture – but also build simulations that played like a movie, allowing us to understand what was going on with direct observation.”

Basically, AI allows a transition toward direct observation to determine which hypotheses are right and what targets are relevant with a better understanding of the mechanisms at work – and according to Harhen, that’s one of the main challenges in this space.

“The hard part is determining the mechanism of action. In my opinion, you need to start with a hypothesis or use all available data to generate one. That’s what AI allows us to do today – combining the data in a logical way to come out with the right hypothesis.”

This is made possible by what we call large quantitative models (LQMs), which we hear less about than large language models (LLMs) that are attracting a lot of hype because of their capacity to ingest vast amounts of text data.

But in biotech and more specifically in the Alzheimer’s space, LQMs have an edge over LLMs. “Quantitative AI – mathematics, physics, molecules – isn’t based on words, and LLMs can’t solve these problems. This is where LQMs come in. LQMs, trained through mathematical models, enable us to understand the physics behind molecular interactions and how molecules behave with their targets,” said Harhen.

This makes LQMs extremely versatile and they undoubtedly hold potential far beyond Alzheimer’s and neurodegenerative diseases. Harhen also pointed out that because LQMs are based on physics, they don’t hallucinate; they’re grounded in the laws of reality.

“I like to think of the AI landscape as a bell curve, with many popular, useful tools falling in the middle. With LQMs, we focus on the edge cases where most of these common tools aren’t sufficient — like in Alzheimer’s,” said Harhen.

AI’s capabilities in drug discovery extend beyond simply screening compounds. AI-based platforms can quickly simulate molecular interactions, assess blood-brain barrier permeability, and predict adverse effects, significantly reducing the time and cost traditionally required.

Beyond SandboxAQ’s simulations, for instance, Exscientia’s AI platform, Centaur Chemist, uses predictive modeling to streamline the selection of candidate compounds. This platform has already brought three AI-designed drug candidates to clinical testing, including DSP-0038 for Alzheimer’s psychosis, which targets serotonin receptors to help alleviate behavioral symptoms associated with the disease.

In partnership with Cambridge University, Insilico Medicine also uses its AI model, PandaOmics, to target proteins implicated in Alzheimer’s disease through a process known as “protein phase separation.” They identified therapeutic targets like MARCKS, CAMKK2, and p62 – proteins likely involved in Alzheimer’s progression due to their tendency to form abnormal protein aggregates.

While no AI drug candidates for Alzheimer’s have completed the full clinical journey yet, it might only be a matter of time until new solutions are brought to patients by LQMs. And what if the next big thing in the Alzheimer’s space was already on the market but we simply didn’t know it held such potential?

Repurposing already approved drugs with AI

Discovering a new drug isn’t the only way to bring novel solutions to patients. “We’ve seen with GLP-1 agonists that they were initially developed for diabetes, but then we discovered a beneficial side effect for weight loss. Instead of competing over a single treatment, companies are now pursuing various indications in parallel, expanding the market. drug repurposing is becoming a big area and I think we’ll see something similar happen with Alzheimer’s treatments,” said Harhen.

Harhen mentioned Every Cure but it’s not the only company aiming to find the next big thing in “old” drugs. Harvard’s Drug Repurposing in Alzheimer’s Disease (DRIAD) framework uses machine learning to screen and identify existing drugs that could potentially treat Alzheimer’s by repurposing them for neuroprotection.

The DRIAD platform has screened numerous anti-inflammatory and neuroprotective drugs, prioritizing those that affect pathways relevant to Alzheimer’s pathology. The DRIAD team applied the screening method to about 80 potential candidates and listed the most promising ones. According to the AI model and the scientists operating it, Janus Kinase (JAK) inhibitors could hold serious potential in the Alzheimer’s space.

Another example is the Dream (Drug Repurposing for Effective Alzheimer’s Medicines) study. Led by the National Institute on Aging (NIA), this study aims to identify and validate drugs initially approved for other conditions but showing promise against Alzheimer’s. Researchers identified 35 FDA-approved drugs that target 20 metabolic pathways associated with Alzheimer’s, narrowing it down to 15 candidates for further analysis. If these drugs prove effective, they may offer quicker and more affordable treatment options for Alzheimer’s compared to traditional drug development methods.

This is another promising AI application but the revolutionary tool still faces some challenges in the neurodegenerative field and more specifically in Alzheimer’s.

What’s missing for AI to find the next big Alzheimer’s treatment?

AI is not just hype but it is possible the hype raised unreasonable expectations. To allow AI to have the transformative impact it promised in Alzheimer’s disease, issues still need to be addressed.

“One challenge is data alimentation; we don’t have that much. It is a very common challenge in AI – there is not enough data, it’s not in the same place, and it’s not organized. The other challenge is that it’s not generalizable. When you look at genetic diversity, the differences are so small they are extremely hard to pick up,” noted Harhen.

AI’s effectiveness in Alzheimer’s research is hindered by the limited availability of high-quality, standardized data. Datasets are often small, heterogeneous, or collected under varying conditions, which complicates training AI models that require reliable and diverse data to generalize effectively across populations.

However, Haren explained that this lack of data isn’t an issue in all AI applications for Alzheimer’s. “For the drug discovery process, we don’t need to ingest a third party’s data for training because we use the data from molecule-to-molecule interactions and that is synthetically generatable. However, when you do need data, I would advise carefully scrutinizing which partner to go with to have the most ethically collected data.”

Beyond the technical limitations, AI can still be a sensitive topic. “I think there is still distrust regarding AI because of explainability. And it’s exacerbated in clinical settings as health is low-trust topic.”

Indeed, there are still concerns about the explainability of how AI reaches its conclusions. For now, AI is like a black box where the intricate mechanisms are quite opaque, and experts are progressively trying to make more sense out of it. Besides the black box problem, data collection is very complex, especially when it comes to particularly sensitive data such as medical data.

The first AI-discovered Alzheimer’s treatment has yet to reach the market and patients, and its arrival may be the final step needed for the public to fully trust these tools that are transforming biotech – not just the neurodegenerative landscape.