To make cell characteristics visible to the human eye, even under a microscope, scientists normally use chemicals that can kill the very cells they want to observe.
Dr. Steven Finkbeiner, director and senior investigator at the Gladstone Institutes in San Francisco (pictured above), recently teamed with computer scientists at Google for a groundbreaking new study funded by The ALS Association Neuro Collaborative through ALS Ice Bucket Challenge donations. The research shows that computers can see details in images without using invasive methods.
For example, using artificial intelligence (AI), computers can examine cells that haven’t been treated with chemicals and find a wealth of data that saves researchers time and effort, and the number of features that can be obtained from images is extraordinary. Extensive hours of researchers sitting at the microscope manually analyzing cell images could eventually become a thing of the past.
Using AI, Finkbeiner and the Google computer scientists discovered they could give researchers a way to surpass regular human performance by training a computer using a method called deep learning. A type of machine learning, deep learning involves algorithms that can analyze data, recognize patterns, and make predictions.
Finkbeiner and Google’s work was recently published in the prestigious Cell journal and is one of the first applications of deep learning in biology, which can be applied to ALS research.
“This is going to be transformative,” said Finkbeiner. “Deep learning is going to fundamentally change the way we conduct biomedical science in the future, not only by accelerating discovery, but also by helping find treatments to address major unmet medical needs.”
Finkbeiner developed the Brain Bot an automated microscopy system that can track individual cells over time, by assigning each cell a social security-like label. Thousands of cells can be tracked at once over many months, which generates huge amounts of data. Finding the best way to analyze this data became a challenge.
Finkbeiner discovered that Google needed a biomedical research project that generated sufficient amounts of data to be amenable to deep learning to test AI in the lab. It was a perfect fit.
“We wanted to use our passion for machine learning to solve big problems,” said Philip Nelson, director of engineering at Google Accelerated Science. “A collaboration with Gladstone was an excellent opportunity for us to apply our expanding knowledge of artificial intelligence to help scientists in other fields in a way that could benefit society in a tangible way.”
Applying Deep Learning to Cell Analyses
Finkbeiner and Eric Christiansen, the study’s first author, invented a new deep learning approach called “in silico labeling.” Enabling a computer to find and predict features in images of unlabeled cells, in silico labeling uncovers important information that would otherwise be problematic or impossible for scientists to obtain without disturbing experiments with traditional labeling methods.
“We trained the neural network by showing it two sets of matching images of the same cells; one unlabeled and one with fluorescent labels,” explained Christiansen, software engineer at Google Accelerated Science. “We repeated this process millions of times. Then, when we presented the network with an unlabeled image it had never seen, it could accurately predict where the fluorescent labels belong.”
This technique successfully identified cell nuclei (the cell’s powerhouse), cell-type in a mixture of cells (to distinguish a neuron versus surrounding support cells), cell health (alive versus a dead cell), and types of subcellular structures, such as a dendrite versus an axon (see below).
“The more the model has learned, the less data it needs to learn a new similar task,” said Nelson. “This kind of transfer learning—where a network applies what it’s learned on some types of images to entirely new types— has been a long-standing challenge in AI, and we’re excited to have gotten it working so well here. By applying previous lessons to new tasks, our network can continue to improve and make accurate predictions on even more data than we measured in this study.”
“Bringing this technology to biologists is such an important goal,” added Finkbeiner.
Potential Significance for ALS Research
In his laboratory, Finkbeiner is trying to find new ways to diagnose and treat neurodegenerative disorders, including ALS.
This AI deep learning system has many applications to better understand ALS. For example, researchers could classify the different types of ALS, which would have significant implications – from changing how researchers study the disease to altering the way clinical trials are conducted. Using this new method, researchers could stratify people with certain disease characteristics to help improve clinical trial outcomes.
Also, with the increased use of induced pluripotent stem cells, cells that can be programmed into motor neurons (the cells that die in ALS) from individuals with ALS, researchers could match a person’s own cells with their clinical information. Deep learning could be applied to find relationships between their symptoms and their cell pathology. Ultimately, this method could help identify subgroups of people with ALS with similar cell features and match them to the most effective therapy to treat their disease.
“The application of AI to further explore novel targets and biomarkers is very exciting. We are pleased to support the innovative work of Dr. Finkbeiner and his team and their exciting partnerships as a result of his efforts,” stated Dr. Lucie Bruijn, chief scientist of The ALS Association.
To read the news announcement about this study from Gladstone, click here.
Paper citation
Eric M. Christiansen, et al. In Silico Labeling: Predicting Fluorescent Labels in Unlabeled Images. April 19, 2018. Cell. 173(3): 792-803.
Join the conversation. Please comment below.