Video Games Help Model Brain’s Neurons


That is far more than professional neuron analysts at places like the Allen Institute can ever hope to handle. Enlisting novices through a game like “Mozak” helps with the task. But the real goal is to use high-quality neuron reconstructions done by humans — the more the better — to train computers to do the job more accurately than they can now.


A still from the “Mozak” video game, which uses common tricks of the medium — points, leveling up and leader boards — to encourage players to participate.

“If we create a corpus of thousands of these neurons, the chances are the computational methods will become better,” Dr. Popovic´ said. “You can see this as symbiosis — computers learn from what people are doing and they do better.”

“Mozak” has helped the Allen Institute increase the number of neuron reconstructions from 2.33 a week that a team of professional analysts were doing on their own, to 8.3 reconstructions a week, said Staci Sorensen, senior manager of morphology at the Allen Institute. That reflects both the contributions of players and the productivity benefits of an internal version of “Mozak” that the institute’s professionals now use.

About 200 people a day play “Mozak” now, and with more players will come more reconstructions.

“I’m hoping ultimately for an order of magnitude, a factor of 10, acceleration,” said Christof Koch, president and chief scientific officer of the Allen Institute. “How big is the community willing to do this? Is it 20 oddballs or thousands?”

“Mozak” is the latest in a growing array of citizen science initiatives with the dual aims of using the wisdom of crowds to tackle complex problems and engaging the public in science. There are public efforts to count bird populations to help scientists understand the effects of climate change. An effort called Galaxy Zoo aids the work of astronomers by getting the masses to classify the shapes of galaxies from telescope imagery.

In 2015, the Obama administration sought to build awareness of citizen science projects through a forum at the White House, creating a portal to catalog them. “Foldit,” Dr. Popović´’s earlier effort, used game mechanics to become one of the most successful citizen science projects, attracting nearly a million players over its lifetime. Since then, similar games have found followings, like “EyeWire,” which uses players to map retinal neurons to help unlock the mysteries of vision.

The spark for “Mozak” came from Jane Roskams, a professor of neuroscience at the University of British Columbia and former executive director for the Allen Institute, when she attended a White House workshop on brain science initiatives a couple of years ago. After meeting with Dr. Popović back in Seattle, the two applied for and received a National Science Foundation grant that they used to develop the game.


Zoran Popovic in his lab at the University of Washington in Seattle.

Stuart Isett for The New York Times

“There are some clear opportunities for really engaging a much broader public audience for tackling some of these big data challenges,” Dr. Roskams said. “Most of the time experts think they need to turn to experts.”

Playing “Mozak” doesn’t offer the slick graphics and sugar rush of matching gems in “Candy Crush.” The game — the name of which means “brain” in Serbo-Croatian, Dr. Popović’s mother tongue — presents players with images of mouse and human neurons captured on powerful microscopes by the Allen Institute in its laboratories.

The neurons in the images, their dendrites and axons emanating from their somas, look like spiders with dozens of legs stretched to the breaking point in all directions. The images are grainy, with dandruff-like flakes floating around that are not part of the neurons. Computers just are not very good yet at constructing 3-D models out of all that noise.

But humans, if they are focused and patient, can be pretty adept at it. “Mozak” players ring up points as they trace the structure of neurons, then advance to a new neuron once they have finished reconstructing one. When multiple people trace the same 3-D models, there is a good chance they are accurate.

Bob Bondi, a retired software developer in Bend, Ore., has been playing “Mozak,” initially logging as much as 10 hours a day but has now tapered back to about three. Mr. Bondi likes the idea of contributing to science by playing the game (players who put in a serious effort are credited for their work).

A chat room allows players to commiserate with each other. Researchers chime in with advice at times. Mr. Bondi said “Mozak” also relaxed him.

“If I have a late night and I’m just not interested in listening to television, I do it to get my eyes tired to go to sleep,” he said. “It’s meditative in a way.”

Continue reading the main story

Source link