Traditional biology has relied on disruption, on breaking systems to see what fails as a clue to how it operates. Today, advanced imaging technologies and AI are changing how biology can be done. With these tools, it is possible to eavesdrop on biology and watch it unfold in real time, from individual molecules to organism-wide systems. Close observation of living systems is not only more efficient than breaking them, it is also more revealing about the individual parts and how they work together to form a whole.
Molecular interactions happen over nanometers and nanoseconds, while humans span meters tall and live for decades. At Biohub, and specifically with our work in imaging, our goal is to develop tools with great enough dynamic range to see biological events across that range of time and space. To do so, we are developing a suite of technologies that will allow us to span a billion-fold in time scale and a billion-fold in spatial scale. We are making the invisible visible.
Here are the five key reasons why I believe AI and imaging are on the precipice of transforming biological research:
1. The hardware and software infrastructure are available
Imaging is at a sweet spot where we can steal liberally from tools developed by the silicon industry and others — tools with impossibly good performance. For example, we now have laser phase plates for use in transmission electron microscopy with mirrors that are several-fold better than the mirrors used at the Laser Interferometer Gravitational-Wave Observatory (LIGO), which already uses mirrors that were remarkably better than I thought was possible. Our physical tools have gotten dramatically better, thanks to fields outside of biology.
Computational tools have also been refined so dramatically that scientists are able to extract knowledge out of huge datasets that wouldn’t have been possible before. A few years ago, I talked about handling big data in biology with someone who did informatics for LIGO and the Large Hadron Collider (LHC). He just started laughing, because the data sizes for LIGO and the LHC are orders of magnitude bigger. The work on wrangling big data has already been fought and partially won by others. Today, just as Google Earth can scale from a front yard to a view of the Earth, there are routine tools, called data pyramids, for doing that same jumping of amazing spatial scales within datasets. Every computer tool I can imagine has come to pass and is allowing us to do things we always thought were either impossible or impractical.
Biologists can now take advantage of this amazing hardware and software infrastructure, especially when powered by AI tools. With AI, computer tools, mechanistic models, and theory, we’ve got the ability to do in biology what applied physics, theoretical physics, and engineering did so fruitfully a century ago — building and testing large-scale approaches to hard problems in the field. We now have the ability to extract truly meaningful knowledge from a wealth of biological data.
2. Progress is moving fast
Progress is going to be fast at interfaces between fields. Let me share a few examples. For decades, laser phase plates for transmission electron microscopy were talked about as a theoretical possibility, and now one of our imaging grantees and our amazing imaging scientists have built them by combining the best-in-class mechanics, electronics, lasers, and more. This tool is allowing us to get more contrast for the same amount of damage to the specimen. We’ve also developed a set of cryotechnologies, like cryo block phase imaging, that enable us to get near-atomic level insights in cell-scaled volumes.
At Biohub we’re also building first-of-their-kind tools for computational imaging. Our bioengineering team makes custom, robust instruments that accelerate the pace of discovery. The computational microscopy team builds microscopes with new imaging modalities that work like cellular paint-by-numbers — they detect what’s there, even if it isn’t labeled. We’re also refining plans to make it possible to analyze billions of cells per week.
Thanks to this combination of expertise across fields, scientists are now imaging better and faster than we ever did before, while acquiring and interpreting the data in real time.
3. The future is n-dimensional
At a team meeting, I brought up how we were planning to do five-dimensional imaging: three dimensions in space, one in time, and one in molecular species. Everybody in the room chimed in to add more dimensions: They said that we needed to add multiple scales — from atomic explanations of molecular interactions to cellular compartments to cells to tissues to organs — and each added a dimension. They said we needed to add predictive capabilities for the future, so they added yet another dimension.
Now, I’ve started saying “n-dimensional,” because no matter what number I say, it’s too small for where this technology is going. That’s what’s so amazing and exciting: We’re brainstorming tools to image the behavior of the system and to predict what’s going to happen in the future. If we can do that, it will open up a huge space in therapeutics and predictive medicine — the ability to visualize a system and predict how it will respond to a virus, an injury, or a therapeutic.
4. AI and experimentalists are working together in a tight, fast loop
What’s so exciting about right now is that AI impacts every stage of the scientific loop: data collecting, data interpretation, modeling, and more. With AI tools, that loop is getting smaller and faster all the time.
We’re also putting the experimentalist in the loop at Biohub. Under most current methods, experimentalists collect data, then throw it over a wall to someone who analyzes it. Most of the time, the person analyzing the data is saying, “If only they had done the right experiment,” and simply does their best with the data they were given. Because of this divided, error-prone process, progress has been slowed, and you can even find papers that list the same protein by different names within the same paper, without pointing out it is the same protein.
We have the AI modelers tell us what data they really need, then we develop tools to get that data. This puts the lab in the loop to make AI modeling faster, better, and more efficient. With new technologies, experimental work is moving as fast as the AI — and we all know that the AI is moving incredibly fast. We’re on the verge of this really fast, virtuous cycle that is going to dramatically increase the speed of discovery.
5. New tools will be interoperable
We’re building a suite of tools that talk to one another. Rather than building the world’s best electron microscope on its own, it has to be the world’s best electron microscope that talks to the world’s best light microscope, which talks to the world’s best next technology. We need to build these technologies so that they’re designed from the start to interact with one another. With that, we’ll achieve new visual resolution that gives us a window into processes that were once invisible, enabling discovery at the pace of life itself.
Across our scientific grand challenges, our teams are pursuing the most informative or most rate-limiting challenges in biology in order to have the greatest impact on human health and disease. That’s what makes me excited: If we see a hard problem, everybody runs to it. We want to solve the unknown. We aim to make the invisible visible. And now is the time to do it.
Scott Fraser is president of imaging and the head of Biohub, San Francisco, where he leads a multidisciplinary team combining single-cell and spatial transcriptomics, high-throughput screens for cell state changes, and cryo-ET imaging of cell signaling complexes to provide multi-scale, multimodal dynamic imaging data.