The Madrona Venture Group recently awarded the Madrona prize to the UW CSEEmbarker project team. The Embarker project uses machine learning to identify disease markers of Alzheimer’s that can help researchers develop treatments.
Alzheimer’s is a progressive brain disorder and the sixth leading cause of death in the United States. There is no cure for Alzheimer’s, and no treatment can prevent its progression. Existing treatments can only treat symptoms of Alzheimer’s such as memory loss and disorientation. In Alzheimer’s patients, abnormal deposits of proteins form in the brain, neurons stop functioning, and brain tissue begins to shrink.
“This is a computational method we developed to solve a biological problem,” UW CSE graduate research assistant Safiye Celik said.
The Embarker project explores the relationship between gene expression levels and Alzheimer’s through machine learning analysis.
“This project combines cutting edge machine learning and big data mining,” UW CSE associate professor and Embarker project Principal Investigator Su-In Lee said.
Genes encode proteins that dictate cell function. Only a fraction of the genes in a cell are expressed at a time. By exploring gene expression, researchers can better understand what genes are involved in the brain abnormalities that cause Alzheimer’s.
“The data measures activity levels of all human genes,” Lee said. “We can eventually find a small set of genes that are important in Alzheimer’s progression.”
There are an estimated 30,000 genes in the human genome. The embarker project combined gene data from patients with Alzheimer’s with knowledge from existing literature about the disorder to get reliable results.
“Our methods are specifically made to make use of a small amount of data cleverly,” Celik said.
Machine learning for understanding gene expression has also been used for diseases like cancer. However, Alzheimer’s is unique because of the limited amount of tissue data available. With cancer, it is common for patients to have tumors removed, which enables researchers to obtain biopsy data from tissues. With Alzheimer’s, tissue cannot be collected from patients until after death because the disorder affects the whole brain.
The Embarker project combines more prior knowledge to better work with this scarcity of patient tissue data.
“Machine learning has the power to fuse multiple data sets so we move in the direction of reliable results,” Lee said.
“If you rely just on the data itself, you end up with a lot of false positives,” Celik said.
By using machine learning to understand gene activity for Alzheimer’s, the project is able to identify key targets for gene suppression. Preventing the expression of certain genes may eventually allow researchers to cure Alzheimer’s or slow its progression.
The researchers use C. Elegans, a type of roundworm that can exhibit Alzheimer’s symptoms, to test suppressing genes suggested by the machine learning algorithm. From there, they observe the worms to see if their condition is improved.
“If we can make these worms less sick, that can be a starting point for developing therapeutics,” Lee said.
So far, this testing has produced positive results. With further research and understanding, this may lead to an effective gene suppression approach for treating Alzheimer’s.
This research is an interdisciplinary effort between computer scientists, doctors, and biologists. With more knowledge about biological mechanisms and gene expression, researchers may be able to finetune the Embarker project to more accurately understand Alzheimer’s as well as apply the algorithm to other diseases.
“AI is not just a tool for doctors, but also something they can contribute to by helping computer scientists,” Celik said. “Without them, we can’t ask the good questions.”
