A global team of Alzheimer’s researchers, drugmakers, and foundations has released three free AI tools to speed the search for treatments. One of them digs through the failed experiments that scientists almost never publish.
More than 99% of Alzheimer’s drug candidates fail in clinical trials, and most of those failures never reach print.
The tools aim to keep labs from repeating dead ends and to surface leads hidden in unreadable mountains of data.
Why Alzheimer’s drugs fail
The numbers behind Alzheimer’s drug failure are bleak. One analysis of trials run from 2002 to 2012 found that out of 244 experimental compounds, exactly one reached approval.
Decades of work have produced a mountain of knowledge. But it sits scattered across millions of papers, huge datasets, and private files that never see daylight. No single researcher can hold it all in their head.
That gap is what Randall J. Bateman set out to close. He is a neurologist at the Washington University School of Medicine in St. Louis (WashU Medicine).
Bateman founded and directs C-BRAIN, a 17-member consortium that led the tools’ release at a major Alzheimer’s conference in London.
A new kind of partnership
Bateman has done this before. In 2011, he pulled ten competing drug companies into a shared consortium.
Together they tested prevention drugs in families with an inherited form of the disease – an unusual truce among rivals. The new effort widens that idea to cover the whole field.
The idea behind the project is that AI can recognize patterns across enormous amounts of data that no individual researcher could ever absorb.
“The brain is immensely complex, but artificial intelligence inspired by the human brain can find relationships within massive amounts of data that a single human mind simply cannot hold,” said Bateman.
Looking beyond published studies
The tool drawing the most attention pulls from what the consortium calls dark data – the unpublished results and dead-end experiments that never make it into a journal. Most science that does not work is simply never reported.
Journals favor positive findings. Negative outcomes pile up in filing cabinets and on hard drives, unseen. One study of drug trials found that positive results were published far more often, and roughly a year sooner, than results that showed nothing worked.
A buried failure still carries information. These negative results can warn the next lab away from a dead end, but only if someone knows the failure happened.
When no one does, teams spend years and millions rediscovering what others already learned.
Avoiding repeated mistakes
The cost of chasing dead ends is not hypothetical. For years, most Alzheimer’s drug programs went after the same suspect – the sticky amyloid plaques that build up in the brain.
They failed repeatedly, at enormous expense. Shared records of those failures could help the field change course sooner.
The Dark Data Analyzer gathers these buried results from academic labs and drug companies in the consortium and makes them searchable.
Its logic is plain. A researcher planning an experiment can check whether someone already ran it and watched it fail.
Open by design
Bateman insists the whole system stay open. “It is antithetical to science that we would develop AI tools that function as an uninterpretable black box,” said Bateman. Anyone can read the code, test it, and find its flaws.
Openness runs into a wall with drug company data, which is guarded and proprietary. The consortium’s answer is a federated design. The AI analyzes data where it is stored, so confidential files never have to be shared.
That design is meant to unlock data companies would never publish. One review estimated that around 60% of trials with disappointing drug results go unpublished – a habit that wastes money and repeats harm. Pulling that material into a shared tool is the point.
Two other tools round out the set. One combs the research literature, helping scientists weigh ideas faster than reading by hand allows.
The other, called Reviewer Three, gives the kind of critical feedback a journal reviewer offers – but for grant proposals and study designs. Human scientists stay in charge of every step.
The tools were built with computing power from a federally funded AI research program and coded by a small in-house team of AI scientists. A public demonstration of what they can do sits on the consortium’s website.
Putting AI to work
What is new is not a discovery but a set of tools, now in researchers’ hands and free to any approved lab working on brain disease.
For the first time, a failed experiment inside one drug company can steer another scientist away from repeating the same mistake.
If the approach works, the payoff is time. Tens of millions of people worldwide live with dementia and wait on treatments. Every failed trial that stays hidden costs researchers years they cannot spare.
Bateman ties the whole venture back to the people it is meant to help. “Aligning drug developers, philanthropic and patient-advocacy groups, and researchers and doctors who have fought these diseases for decades, and giving them these AI tools, is how we deliver on our promise to patients,” he said.
—–
Like what you read? Subscribe to our newsletter for engaging articles, exclusive content, and the latest updates.
Check us out on EarthSnap, a free app brought to you by Eric Ralls and Earth.com.
—–
