Researchers retreat to bucolic surroundings for brain cell-mapping competition
EDITOR'S NOTE: Instead of awarding a top prize, DIADEM judges have chosen to distribute the $75,000 prize money among four of the five teams in the competition. The winning teams are listed here.
Nestled in the bucolic hills of Virginia, the Howard Hughes Medical Institute’s Janelia Farm campus easily lulls visitors into a state of relaxation. But right now, teams of computer scientists are ignoring the picturesque surroundings. They’re hunkered down in pods, staring at large computer monitors, confabbing with neuroscientists and feverishly competing over the next few days to see whose program can best untangle intricate webs of brain cells.
“This is a landmark event,” contest judge and neuroscientist Kevan Martin of the University of Zurich said in a talk on August 30. “Enormous things are going to come from this, I’m sure.”
In addition to accolades, the winning team will take the $75,000 top prize in this first-ever brain-mapping competition called DIADEM, short for Digital Reconstruction of Axonal and Dendritic Morphology, which runs from August 29 to September 1. Right now, it can take brain scientists an entire month to painstakingly hand draw a single brain cell from microscope images. But the brains behind the DIADEM Challenge thought that computer programs could speed things up.
“Because of the complexity and sheer number of neurons and their connections in the brain, it is very clear that this is not something that can ever be solved manually,” says conference organizer and neuroscientist Giorgio Ascoli of George Mason University in Fairfax, Va. “We need machines to do that.”
To the rescue come computer algorithms designed by 29 computer scientists working in five teams. After having made it through the early stages of the competition, the algorithms are being put to the test in the finals on six real datasets of neuron images generated by leading neuroscientists.
These brain specialists provided images of twisty, complex brain cells from rats, cats, mice and fruit flies. The teams will be judged on two criteria: their programs’ speed and how well their algorithms’ drawings matched a neuroscientist’s hand drawing of the same neurons.
As I wandered between pods of computer scientists and neuroscientists, I overheard quiet conversations: “Why aren’t these dots connected?” and “Move the trace a little left,” and “Why do you think the axon is shaped like that and not like this?”
The DIADEM teams had seen five of the datasets before coming to Janelia Farm, but to spice up the competition, organizers let loose a wild card sixth dataset — an image of a highly branched neuron from a cat’s visual cortex provided by Judith Hirsch of the University of Southern California in Los Angeles.
The neuroscientists who provided the data want to understand how the brain’s organization makes it work. Neurons’ shapes and connections control our ability to appreciate the slope of a grassy hill, move the finger muscle that controls the click of a mouse and smell a cup of espresso (caffeine was abundant at the competition).
Teasing apart brain cells’ connections could lead to a deeper understanding of how neurological disorders, such as schizophrenia, take hold. “There are profound errors in connections in the brains of people with mental disorders,” says conference participant Dennis Glanzman, a neuroscientist at the National Institute of Mental Health in Rockville, Md. Finding these connection mistakes and figuring out when and why the brain cells go wrong could lead to better treatments for neurological disorders.
Ascoli says the competition itself is a real-life experiment: It unites two branches of science that have a lot to teach each other. “Ultimately we are all interested in the same goals, so there is a good spirit and good collegial sense,” he says. But it’s not all sunshine and roses. “At the same time, it’s a competition, so there is some tension on the pitch.”
The winning team will be announced on September 1, and the results, datasets and algorithms will be described in an upcoming special issue of Neuroinformatics.