If you play Eyewire, you’ve probably heard us talk about Neo for what seems like years and years and years. It’s a long time coming here, too! Yet none have been idle.
Princeton University’s Seung Lab has been cranking away for all of these years, alongside partners at Baylor College of Medicine and The Allen Institute. Together, we’re building the necessary infrastructure to support such a monumental reconstruction of roughly 100,000 neurons and 1 billion synapses. We’re nearing the end of Phase 2 of 3 of the public-private partnership that is the IARPA MICrONS program. Up next is the final phase, in which Neo will finally launch to the public.
How far we have come
The quest for the connectome of a cubic millimeter of cortex began when teams at Baylor College of Medicine (BCM) and the Allen Institute for Brain Science (AIBS) selected, prepared, and imaged this large volume. Beginning at BCM, researchers took daily recordings from the same area of the visual cortex while mice watched a variety of movies ranging from custom-designed stimuli to scenes from Hollywood action flicks like Mad Max Fury Road. The BCM team tried to record enough data from each neuron in the cubic millimeter to characterize their unique response properties – the small part of the visual world that they respond to, and their preferred stimulus at that location. After recording was completed at BCM, the mice took a direct flight to Seattle, where the same millimeter cube was processed for electron microscopy.
The imaging at the Allen Institute pumped out an astounding 285,000 images every day for months on end. If you’d like to see some pics from the process, check out this Twitter feed from a scientist on the AIBS team, Dan Bumbarger. While imaging was still taking place, the challenge of getting petabytes of data from Seattle, WA to Princeton, NJ unfolds. The data were nearly sent by physical hard drives. Ultimately, a digital ingestion tool was built to ferry more than a quadrillion voxels (3D pixels) from Seattle to Princeton. A single final electron microscope image from Allen contains between 250,000 and 500,000 pixels on each side, each of which images 640 cubic nanometers.
Upon receipt of images by Seung Lab, they are oriented and aligned. That brings us to today. As I type, the team is aligning like the inverse of a tornado while Princeton tracers are cranking to generate ground truth data that will be used to train the AI that will eventually segment the final Neo dataset.
We are in Seung Lab’s crunch months of Phase 2. For a sense of scale of what is to come in Phase 3: if you’ve seen Eyewirer @kfay’s render (below) that graced the walls of Princeton University as a mural, or perhaps played his beautiful cortex fly-through, then you’ve seen renders from a volume that is 1/1000th of the final size of Neo.
The team has also been working on a new web viewer, annotation tools, and other features that enable web-based collaboration in reconstructing vast neurons from truly gigantic image datasets.
To get a sense of scale, check out a comparison of Eyewire dataset/cells Vs that of a test cortical volume from Phase 2. The Eyewire dataset is shown as a white rectangle bottom left. Eyewire is not just smaller, it’s thinner, too. Eyewire is 6.3 Million microns^3 while the final Neo datset will be about 1 Billion micron^3. That’s 150x larger.
We hope you’ll bear with us for a few more months while we continue to work tirelessly to bring Neo to life. We’re beyond thrilled at the implications of this research and the exhilarating possibilities of inviting the world to help map these beautiful, intricate neurons. Keep an eye out for more updates as we draw closer to the Alpha Launch of Neo!
Thanks to Nicholas Turner and Jacob Reimer for helping with this post!