I recently attended a biology conference where, among many other things, I got to see a talk by Dr. Jeff Lichtman of Harvard University on brain connectomics research.
It’s very interesting stuff. He has produced a set of custom equipment that can scan brain tissue (well, any tissue, but he’s interested in brain tissue) at 5x5x30 nm resolution. His super-duper one of a kind electron microscope can at this point scan about 0.3 cubic millimeters in 5 weeks, if I’m not mistaken. It spits out a dataset in the fractional-petabytes range. He’s had one such dataset for a full 3-4 years but is encountering major problems with analysis—tracing cells and fibers over their full path is a very difficult problem. Automatic cell-tracer programs are good enough over the number of slices that makes up a cell-body but utterly fail at identifying things like synaptic vesicles reliably and when tracing fibers over their full lengths. To the point that most of his good data that he showed us has been manually annotated by graduate students and undergrads working in his laboratory. Hence the above link’s mention of gamifying the task to try to crowdsource it.
Interestingly, he described his equipment as a ‘tissue observatory’. He thinks that neuroscience should take a page from astronomy and just see what the heck is out there. He thinks they are trying to make detailed hypotheses about function and structure and everything else on far far too little data right now and we need a lot more data on the actual structures before we can be confident about much about them other than the mere fact that they correspond to function. To the point that during his talk, when showing the 30 micron wide completely-perfectly-annotated chunk of his first dataset that he has published much analysis on the thousands of synapses of (something like 0.01% of his first raw dataset he has had for years) he showed the exploded figure of hundreds of cellular fibers and the table of dozens of parameters for thousands of synapses and said “And here it is… incredibly beautiful and so far totally useless.” His point being he wants to annotate more of it, and use it as a base to make actual informed inferences and hypotheses about connection formation and network structure. He also notes that his datasets cannot tell apart different neurotransmitter producing cells, gene expression, or the presence of different kinds of proteins pre or post synaptically.
He (knowingly for humor and exasperation’s sake) overstates the case there about ‘uselessness’ even at current levels of annotation—you can focus in on pieces of the dataset and annotate it for your own purposes. Someone else at the conference presented work in which they took his dataset (which is, after all, revolutionary in terms of the sheer amount of fine 3-dimensional data it has on the structure of so many different cell types in their normal living context) and figured out longstanding questions about the topology of certain intracellular structures that have held for quite a long time. He already has interesting statistical information about the connections between the cells in his pefectly-annotated segment of data, showing that if two cells are connected they tend to be connected in multiple places. He also appears to have found cell types in his data he had no idea existed and he still doesn’t know what they are, and noted that the big spine-based synapses that have been well-studied so far only represented less than a third of the synapses in the perfectly-annotated chunk. There’s apparently other people lining up to use his equpment too and if I recall correctly he said someone is hoping do an entire fruit-fly brain, much like was mentioned in the above link.
He thinks that neuroscience should take a page from astronomy and just see what the heck is out there. He thinks they are trying to make detailed hypotheses about function and structure and everything else on far far too little data right now and we need a lot more data on the actual structures before we can be confident about much about them other than the mere fact that they correspond to function.
I recently attended a biology conference where, among many other things, I got to see a talk by Dr. Jeff Lichtman of Harvard University on brain connectomics research.
It’s very interesting stuff. He has produced a set of custom equipment that can scan brain tissue (well, any tissue, but he’s interested in brain tissue) at 5x5x30 nm resolution. His super-duper one of a kind electron microscope can at this point scan about 0.3 cubic millimeters in 5 weeks, if I’m not mistaken. It spits out a dataset in the fractional-petabytes range. He’s had one such dataset for a full 3-4 years but is encountering major problems with analysis—tracing cells and fibers over their full path is a very difficult problem. Automatic cell-tracer programs are good enough over the number of slices that makes up a cell-body but utterly fail at identifying things like synaptic vesicles reliably and when tracing fibers over their full lengths. To the point that most of his good data that he showed us has been manually annotated by graduate students and undergrads working in his laboratory. Hence the above link’s mention of gamifying the task to try to crowdsource it.
Interestingly, he described his equipment as a ‘tissue observatory’. He thinks that neuroscience should take a page from astronomy and just see what the heck is out there. He thinks they are trying to make detailed hypotheses about function and structure and everything else on far far too little data right now and we need a lot more data on the actual structures before we can be confident about much about them other than the mere fact that they correspond to function. To the point that during his talk, when showing the 30 micron wide completely-perfectly-annotated chunk of his first dataset that he has published much analysis on the thousands of synapses of (something like 0.01% of his first raw dataset he has had for years) he showed the exploded figure of hundreds of cellular fibers and the table of dozens of parameters for thousands of synapses and said “And here it is… incredibly beautiful and so far totally useless.” His point being he wants to annotate more of it, and use it as a base to make actual informed inferences and hypotheses about connection formation and network structure. He also notes that his datasets cannot tell apart different neurotransmitter producing cells, gene expression, or the presence of different kinds of proteins pre or post synaptically.
He (knowingly for humor and exasperation’s sake) overstates the case there about ‘uselessness’ even at current levels of annotation—you can focus in on pieces of the dataset and annotate it for your own purposes. Someone else at the conference presented work in which they took his dataset (which is, after all, revolutionary in terms of the sheer amount of fine 3-dimensional data it has on the structure of so many different cell types in their normal living context) and figured out longstanding questions about the topology of certain intracellular structures that have held for quite a long time. He already has interesting statistical information about the connections between the cells in his pefectly-annotated segment of data, showing that if two cells are connected they tend to be connected in multiple places. He also appears to have found cell types in his data he had no idea existed and he still doesn’t know what they are, and noted that the big spine-based synapses that have been well-studied so far only represented less than a third of the synapses in the perfectly-annotated chunk. There’s apparently other people lining up to use his equpment too and if I recall correctly he said someone is hoping do an entire fruit-fly brain, much like was mentioned in the above link.
Upvoted especially for this.