What are the most transformative neuroscience tools and technologies developed in the past five years?

“Modern machine-learning methods all of a sudden have become so usable they can all be applied to neuroscience in pretty short order. PyTorch in particular seems widely used in neuroscience, and it is relatively sudden that anyone/everyone could be training big models and applying them to all kinds of problems in science. Both things that were primarily useful at reducing labor, like pose tracking in animals or cell segmentation in images, and also bigger, conceptual models of learning, cognition, etc. The barrier to entry suddenly dropped and it’s still kind of remarkable to me how low the barrier to entry became. You can generate a few lines of code and be off to do things that we couldn’t solve for years previously.”
—Joshua Dudman, senior group leader, Janelia Research Campus, Howard Hughes Medical Institute

“I think there’s been a trend towards building more quantitatively powerful models of the brain. This is a place where the AI models come into play, where we can now make use those kinds of models, particularly neural network models, to make predictions about brain activity that have much greater predictive power than any of the models that people had built previously.”
—Samuel Gershman, professor of psychology, Harvard University

“Computational tools such as Zebra for so-called joint embedding of neural signals and behaviors that really allow us to understand how neural signals relate to behavior.”
—Johannes Kohl, group leader, Francis Crick Institute

“Geometric/topological machine learning theory. This area of mathematical development, I believe, is key to paving the way to developing a much richer picture not only of the full richness of what brains can represent and learn, but also give structural hypotheses and true mechanisms at a mathematical level of how it is that particular structures of knowledge and representation come about and enable affordances in the world.”
—Maxine Levesque, M.D.-Ph.D. candidate, University of California, San Francisco; platform architect, Astera Institute

“The development of things like more sophisticated statistical models, machine-learning models to allow people to take these large datasets and try and make sense of them. Now they [technologies] have led to a problem that sometimes you don’t actually know what the answer really is to get from machine learning or AI. It just gives you an answer. So that has led to another thing, which was anticipated, is that it focused so much on the data that we sort of lost the touch. What is the data actually telling us? How are our theories developing at the same time? So, the technologies were much faster than our understanding and certain technology much, much faster than corresponding theories of brain function and dysfunction.”
—Randy McIntosh, director, Institute for Neuroscience and Neurotechnology, Simon Fraser University

“There has been, I think, a huge breakthrough in computational analysis of behavior, computational approaches to analyzing data. I think the most recent potential excitement is in AI approaches to understanding data and putting together those sorts of things.”
—Jason Shepherd, professor of neurobiology, University of Utah

“Deep-learning-based tools to extract information from nonlinear complex neural data that is behaviorally relevant.”
—Jimmy Singh, postdoctoral research associate, University of Utah

“Mathematical techniques for understanding the activity of large collections of neurons are incomplete yet, but we’d be nowhere without them. They can be overinterpreted.”
—Michael Stryker, professor of physiology, University of California, San Francisco

“Automating behavior tracking of all animals with many cameras at once. That was happening more than five years ago, but we’ve seen rapidly accelerating technology.”
—Bing Wen Brunton, professor of biology, University of Washington

“DeepLabCut has been a big deal. That’s the machine-learning technology that allows people to track the movements of animals with regular video cameras. That’s been very transformative because now nonspecialized labs that aren’t serious machine-learning labs can do behavior experiments and have quantitative data for it.”
—Gregory W. Schwartz, professor of ophthalmology, neuroscience and neurobiology, Feinberg School of Medicine, Northwestern University

“Neuropixels probes—they came out and they were immediately usable and everyone started using them. And so, everybody went from, you know, on the order of tens of neurons recorded per session to something like hundreds or close to thousands recorded per session. And that’s just a huge leap and it had really fast uptake. So, I think that was a really big piece of technology.”
—Joshua Dudman, senior group leader, Janelia Research Campus, Howard Hughes Medical Institute

“High-density recording methods—for instance, Neuropixels electrodes—which you can use to record from thousands of neurons in behaving animals.”
—Johannes Kohl, group leader, Francis Crick Institute

“Neuropixels probes: The ease and density of sampling, as well as small size enabling multisite recording, has already enabled cross-region studies that tell us how specific findings are to each area and better assess interregional communication. Voltage dyes, when they really mature, will take these features to the next stage.”
—Samuel Levy, postdoctoral scholar, neurobiology, Stanford University

“A lot of people are using Neuropixels as well, so multi-electrode arrays that have more and more electrodes that you can get into their brains. And that’s another one where standardization and commercialization actually helped. People can just buy a Neuropixels and do a bunch of spike sorting and everything with it, and it’s easy and people can record from hundreds to thousands of cells simultaneously. So that’s made that more accessible in the community.”
—Gregory W. Schwartz, professor of ophthalmology, neuroscience and neurobiology, Feinberg School of Medicine, Northwestern University

“In-vivo probes to target hundreds to thousands of neurons simultaneously with high temporal precision, such as Neuropixels probes.”
—Jimmy Singh, postdoctoral research associate in Christopher Gregg’s lab, University of Utah

“Large-scale recordings, which are getting better and better. It’s increasingly possible to maintain this over long times on the same cells. There have been recordings with high temporal precision, electrically, but we can also follow changes over time in the same cells. That’s quite new.”
—Michael Stryker, professor of physiology, University of California, San Francisco

“The jury remains out on Neuropixels recording and what that can teach us. It may be an unpopular opinion. When I was in graduate school we would have 10 cells, now we have 10,000. That sounds like a lot, but it’s a small fraction of the mammalian brain. Does recording from an order of magnitude more cells teach us something new, or just rigorously confirm what we already knew? I’m skeptical. I’m more excited about the comprehensive recordings of things happening in brain and behavior. Ten thousand cells is a small fraction, but if technology can record from calcium imaging the entire cortical surface, that’s the entirety of the top. That will allow us to do new things.”
—Bing Wen Brunton, professor of biology, University of Washington

“Two- and three-photon imaging in primates, mice and rats. That gives us more depth, high specificity. Transcranial ultrasound to activate tissues, with fMRI, the combination with noninvasive techniques. The ability to modulate circuitry and see changes.”
—Martijn Cloos, associate professor of bioengineering, University of Queensland

“Brain expansion. So, this is the idea that you put in these polymers, you basically get them into tissue and then when they’re exposed to water, they expand. And so, what you can do is you can take tissue and expand it 10 times larger, and this allows you to image things with a standard microscope instead of having to do, say, an electron microscope. That was one thing that was a pretty big deal. But what really got learned there is that you can create this molecular scaffold inside of tissue and you can just scaffold all this stuff to it. And so, all these techniques got better. You can hook up all the RNAs in situ into this embedded polymer, and it’s advancing like all of these techniques as we learn how to work with the chemistry of these polymers better, and so we can infuse them in brains, anchor anything we want to them and get really beautiful measurements of molecular protein, structural, etc. Histology has always been important in neuroscience. That’s pretty well revolutionized histology, so it was a big deal.”
—Joshua Dudman, senior group leader, Janelia Research Campus, Howard Hughes Medical Institute

“Miniscopes, one-photon miniscopes, but also now two-photon miniscopes, which allow us to record calcium dynamics from individually resolved neurons in behaving animals.”
—Johannes Kohl, group leader, Francis Crick Institute

“Imaging in general has been transformative for neuroscience, no question. And some of the big innovation started sort of in the early ’90s with the evolution of MRI, and they just continue to explode in terms of what we can measure with imaging. All the way from, like, white matter, gray matter to now we’re getting to be able to measure calcium flow of cerebral spinal fluid in fast times, protein metabolism, things of that sort. All of those [imaging] techniques really are giving us more detail and being able to break part of the system into smaller and smaller bits but getting the collection of the whole. So, you have the entire brain being measured for blood flow or the entire brain being measured for calcium, for example. So, it’s giving us access to data that we never had even thought of having access to before.”
—Randy McIntosh, director, Institute for Neuroscience and Neurotechnology, Simon Fraser University

“Imaging. It has become more broadly used. This is two-photon imaging for many different applications. Also there have been new approaches to clear tissue to give better fluorescence imaging of the whole tissue; sort of not having to only look in one region or digitally slice it, you get the whole view of the tissue.”
—Steven Proulx, group leader, University of Bern

“The miniaturization of some of the optical components so you can do two-photon microscopy and see cells deep in the brain of a mouse while it’s running around, because they made the microscope 2 grams or something and it can fit on their head. Those parts have been pretty impressive.”
—Gregory W. Schwartz, professor of ophthalmology, neuroscience and neurobiology, Feinberg School of Medicine, Northwestern University

“Everyone is single-cell, single-nuclear. It’s not that impressive. Spatial information is really important. Transcriptomics, where populations reside, in normal, malignant and diseased brains. There are new tools on spatial. We can get more granular with disease and cell states.”
—Benjamin Deneen, professor and Dr. Russell J. and Marian K. Blattner Chair, Baylor College of Medicine

“Single-cell sequencing in the last five years has really led to a lot more knowledge of which types of cells comprise different parts of the CNS and is starting to allow us to better understand, at least in my field, how the brain barriers are arranged and try to elucidate more of the anatomy that has been amazingly unclear still.”
—Steven Proulx, group leader, University of Bern

“There’s of course been a lot of molecular tools developed to look at—the so-called ‘-omics’ approaches like RNA sequencing, epigenetic changes, DNA changes. And that’s created a wealth of data, but I think the challenge there is using these new technologies to understand it.”
—Jason Shepherd, professor of neurobiology, University of Utah

“Single-cell sequencing and sequencing in general. Another transformative tool is what people call ‘spatial transcriptomics,’ which I always like to put in air quotes because it is just multiplexed in situ, which we have been doing forever. So what makes it spatial transcriptomics instead of multiplexed in situ is my opinion, but that is my own opinion. But it is a way to do it; it is just a higher throughput way to do it so that you can look at gene expression in different cell types across distances.”
—Anne West, professor of neurobiology, Duke University School of Medicine

“The fusion of genetics with the ability to monitor changes in neurons and glial cells in real time is fundamentally important to understanding the brain, circuits and behavior.”
—Benjamin Deneen, professor and Dr. Russell J. and Marian K. Blattner Chair, Baylor College of Medicine

“Pathway-specific axon silencing using opsins to suppress vesicle fusion at the synaptic terminal is necessary to understand how information flows between different brain areas. Silencing the somatic compartment of neurons affects many downstream targets simultaneously.”
—Scott Pluta, assistant professor of biological sciences, Purdue University

“Not that it’s been necessarily only [in] the last five years, but the viral technologies keep getting better to deliver different things to the cells. So, you can deliver lots of different elements via AAV viruses. So that allows you to light up different cells, light up parts of different cells, knock out particular components. So, we have a great suite of tools in mice now to do those things.”
—Gregory W. Schwartz, professor of ophthalmology, neuroscience and neurobiology, Feinberg School of Medicine, Northwestern University

“Optogenetic tools that allow you to access and image cells in the intact brain as the animal is behaving have been super informative, both from an observational perspective and also being able to manipulate the circuits with channels and sort of have causal experiments to say, ‘OK, we think this circuit does this in terms of behavior, but now we can actually show that in a causal way.’”
—Jason Shepherd, professor of neurobiology, University of Utah

“The tool that has now been most transformative are enhancer AAVs that enable cell type-specific targeting for recording or manipulation. And all optical perturbation, to be able to target manipulations in real time to identify cells. We’ve been able to replace electrodes with optical perturbation and recording. Optical technology is really quite new. Optogenetics is a decade old. Chemogenetics is about a decade old, perturbing the activity of genetically targeted cell types. Both are slow; they don’t have the sub-millisecond time that electrical recordings have. In combination, they are really powerful.”
—Michael Stryker, professor of physiology, University of California, San Francisco

“Circuit-breaking technologies, the ability to test the function of different circuits in vivo by putting in channel rhodopsin and halo rhodopsin is useful because it allows you to identify certainly at least what a circuit can do. I am always more interested in what a circuit does.”
—Anne West, professor of neurobiology, Duke University School of Medicine