I've been reading about what is preventing us from getting a whole mammalian connectome scanned today. Here are links to the papers I found:
(Hayworth, K. J. (2012). Electron imaging technology for whole brain neural circuit mapping. International Journal of Machine Consciousness, 4(01), 87-108.)
This is Kenneth Hayworth's 2012 paper describing a facility he calls, a connectome observatory. This facility he estimates can be built for about $200 million and will give us random access to any part of the brain. Scanning the entire brain will still take years though.
(Hayworth, K. J. (2012). Electron imaging technology for whole brain neural circuit mapping. International Journal of Machine Consciousness, 4(01), 87-108.)
This paper from 2013 reviews the challenges stopping us from getting whole connectomes. I liked this paper because it described, with pictures, some of the difficulties with reconstructing connectomes from stacks of electron microscope (EM) images. E.g. if the dotted line is the plane at which the brain was sliced, it is "difficult" to tell which of the three configurations of neurons the slice represents.
The paper also has this neat graph showing the time it takes to image a brain versus to reconstruct the connectome from the images. We're talking about a 3 or more orders of magnitude difference.
(Helmstaedter, M. (2013). Cellular-resolution connectomics: challenges of dense neural circuit reconstruction. Nature methods, 10(6), 501-507.)
There's a more recent review of the challenges in acquiring a whole-brain connectome by Shawn Mikula (a contestant in the BPF prize). It seems to be his opinion (p.4, last para.) that we'll be able to close the analysis gap in the previous graph from Haelmstadedter's paper. The main computational challenge he identified is how to validate the reconstructed connectome. He says
(Mikula, S. (2016). Progress towards mammalian whole-brain cellular connectomics. Frontiers in Neuroanatomy, 10, 62.)
Well, it seems like academia is making fast enough progress on studying the connectome. If you want to try your hand at reconstructing (parts of) a connectome from EM images check out the following.
The Eyewire project (videogame where the aim is to reconstruct parts of a connectome) has a Google drive with a bunch of training data i.e. EM images of a mouse retina with the neurons labelled. Here's an example training pair.
The FlyEM group at Janelia farm also publishes a bunch of images and connectome reconstructions from a fly brain. And this powerpoint presentation talks about the analysis problems they're facing as of September 2016. FYI a fly brain may be harder to analyze since it's neurites are smaller than those of a human. But hey, if you can efficiently extract a fly's connectome, you should be able to do the same for a mammal, right?
The paper also has this neat graph showing the time it takes to image a brain versus to reconstruct the connectome from the images. We're talking about a 3 or more orders of magnitude difference.
(Helmstaedter, M. (2013). Cellular-resolution connectomics: challenges of dense neural circuit reconstruction. Nature methods, 10(6), 501-507.)
There's a more recent review of the challenges in acquiring a whole-brain connectome by Shawn Mikula (a contestant in the BPF prize). It seems to be his opinion (p.4, last para.) that we'll be able to close the analysis gap in the previous graph from Haelmstadedter's paper. The main computational challenge he identified is how to validate the reconstructed connectome. He says
While consensus of redundantly- and manuallyh-traced neurites and annotated synapses is one appraoch, it relies on model-based assumptions.I'm not sure what "model-based assumptions" here means but I'm guessing he's saying that a bunch of human annotators look at a reconstruction and agree, "this looks right." Some other (better) options are:
- Imaging a few parts of the brain at a much higher resolution (e.g. 2x2x2nm instead of 10x10x10nm) and checking that all the relevant features visible at the higher resolution also appear at the lower resolution.
- Trace a few neurons unambiguously (e.g. by making them fluoresce) and then compare it with the tracing from the electron microscope images.
(Mikula, S. (2016). Progress towards mammalian whole-brain cellular connectomics. Frontiers in Neuroanatomy, 10, 62.)
Well, it seems like academia is making fast enough progress on studying the connectome. If you want to try your hand at reconstructing (parts of) a connectome from EM images check out the following.
The Eyewire project (videogame where the aim is to reconstruct parts of a connectome) has a Google drive with a bunch of training data i.e. EM images of a mouse retina with the neurons labelled. Here's an example training pair.
The FlyEM group at Janelia farm also publishes a bunch of images and connectome reconstructions from a fly brain. And this powerpoint presentation talks about the analysis problems they're facing as of September 2016. FYI a fly brain may be harder to analyze since it's neurites are smaller than those of a human. But hey, if you can efficiently extract a fly's connectome, you should be able to do the same for a mammal, right?
