Crowdsourcing Algorithms for Improved ALS Treatments

Data analysis might not be the most glamorous side of medicine, but it can lead to important medical breakthroughs. That’s what happening with ALS clinical practices thanks to a gigantic ALS clinical trial database—and more than 1,000 scientists who joined a crowdsourcing contest to design algorithms that would accurately predict ALS progression.

It’s hard to predict how ALS will progress in a patient, which makes it hard for doctors to choose a good treatment for patients. In a quest to make predictability more reliable, a team of researchers decided to give crowdsourcing a whirl. They pooled together data to form a huge clinical database and set up a challenge for the scientific community: Analyze the data to detect patterns in patient populations—and create an algorithm that will accurately predict how the disease progressed 9 months later for those patients. The challenge resulted in two algorithms reliable enough that they’re now being adapted for clinical practice.

What is ALS?

ALS stands for amyotrophic lateral sclerosis. It’s a terrible, fatal disease that affects the central nervous system. It’s also known as Lou Gehrig’s disease after the famous Yankees baseball player diagnosed with it in the 1930s.

With ALS, patients lose motor neurons, which are the nerve cells that give you the ability to move. If you want to move your arm, for example, impulses from your brain zoom down your spinal cord, traveling via these nerve cells to your arm muscles. Damage to motor neurons mean fuzzier and fuzzier messages from the brain. The nerve cell degeneration leads to weaker muscles, muscle dystrophy, and eventual paralysis, and patients gradually lose their ability to even swallow or breathe. Most patients only live about 3 to 5 years after the first sign of ALS symptoms appear.

The rate at which the disease spreads isn’t the same for all patients, though. There’s variation in when symptoms show up or develop, and some people’s bodies can fight off the conditions longer. The plight for doctors lies in trying to assign effective treatments, without knowing how the disease will manifest itself for each patient. Researchers wanted to unearth patterns in the data that would make diagnosis and treatment more precise.

Bioinformatics Steps in...

Enter bioinformatics. Bioinformatics is all about gathering and exploring huge amounts of biological data. Usually, it’s genetic information. To take on the challenge of predicting ALS progression, however, researchers needed clinical trial data—and lots of it. The idea was that there had to be some details that hint at how quickly ALS will move. A bigger database would help researchers pinpoint more potential predictors. It would also make it easier to confirm whether specific details forecast a specific result reliably over and over again.

Assembling the massive amount of ALS clinical data—and analyzing it—required teamwork. The two main leads for the project and subsequent paper published in Nature Biotechnology last week made an ideal pairing. Dr. Robert Küffner is a bioinformatician at Ludwig-Maximilians-Universität München in Munich. Dr. Neta Zach is the chief scientific officer at Prize4Life, a non-profit organization that attracts new researchers and encourages ALS research developments with prizes.

Together with a team of international researchers, Prize4Life, the IBM DREAM project, and pharmaceutical companies, they compiled the biggest ALS clinical database that had been built up to that point. The Pooled Resource Open-Access ALS Clinical Trials (PRO-ACT) database contained clinical information from 1822 ALS patients. The data, collected within the first 3 months after patients’ diagnoses, came from a variety of clinical trials. Küffner and his team standardized the data points, then opened PRO-ACT up to challenge participants in late 2012.

The Contest Result

The challenge drew in more than 1000 competitors from around the world who submitted 37 algorithms. To test how reliable each algorithm was, the researchers compared the predicted results to what really happened. The two best algorithms outperformed current models as prediction tools—really exciting news for doctors and patients. The study also revealed some surprising, new information. Blood pressure and levels of uric acid and creatinine in blood turned out to be great benchmarks for tracking the disease’s progress, according to the analyses.

The crazy thing is, not everyone who submitted an algorithm came from an ALS research background. In fact, the prize-winning teams both included researchers with statistics or quantitative modeling backgrounds. What made the algorithms successful was that participants had a good knowledge of methods for analyzing clinical data.

Excited by the success of the first crowdsourcing challenge, Küffner is already planning a second. The PRO-ACT database has expanded to include data from 8,600 patients, nearly five times more than it started with in 2012. The hope is that more data will lead to further breakthroughs for ALS researchers and doctors. With the winning algorithms from the first challenge already being adapted for clinical practice, it’s hard to imagine results from the second one won’t lead to more advances.