Why Maslow’s Hammer is a great approach to biology
Turning everything into a sequencing problem has been amazingly successful. Now let's turn everything into an imaging problem.
As the saying goes, when you have a hammer everything looks like a nail. This mentality, known as Maslow’s hammer, is often thought of as a bad thing, leading people to stick with the familiar tools rather than finding new and better ways to solve problems. But when you have a really powerful hammer, looking for nails can turn out to be an effective strategy.
For the past 15 years or so, massively parallel sequencing has been an astonishingly powerful hammer deployed to address questions in essentially every corner of biology, from the origins of Indo-Europeans to the biogeochemical cycling of organic material in Lake Tanganyika. One surprisingly reliable rule for success has been take your desired biological measurement and turn it into a sequencing problem. Using sequencing, we measure the genomic locations of DNA binding proteins, the level of transcription of genes, the identities and states of cells in a complex tissue, the microbial composition of ecosystems, and the familial relationships among ancient human remains. You can even turn your data storage and retrieval problem into a sequencing task.
This “Maslow’s sequencer” approach to biological measurement became even more effective with the creative and widespread use of synthetic DNA “barcodes” to label just about anything in your assay. Instead of testing things one-at-a-time or in parallel arrays of samples, barcoding lets you pool your perturbations and read out the results in one mix. Among other things, this approach lets us measure what’s happening inside single cells at a scale that would have been unimaginable when I was in grad school.
Given how well treating everything as a sequencing problem has worked, we should be on the lookout for the next biological hammer. My Washington University colleague Willie Buchser makes a good case that it will be imaging. Willie is a neuroscientist who heads up the Functional Imaging for Variant Elucidation (FIVE) core at the McDonnell Genome Institute, and he spends his days thinking of ways to use imaging data as the basis of scalable, functional genomic assays.
Willie gave a talk here this spring in which he argued that there is so much more we can be doing with the imaging data coming out of the new spatial transcriptomics technologies. Given the capacities of the new platforms coming online, Willie argues that we should be thinking more about image-based functional genomics, rather than just the sequence-based approaches that dominate now. He makes a good case that the combination of spatial transcriptomics with machine learning-based image analysis promises to tell much more about what’s going on in the complex environment of a whole tissue. The key to success will probably be the deep learning pipelines that will let you scale up this kind of analysis. Scalability is the reason why sequencers are such good hammers, while image analysis has typically been slow, requiring frequent input from human experts. If you overcome that slow step and can process complex histological images with reliable, automated tools, then you’re off to the races and ready to think about how to turn every measurement problem into an imaging problem.
In the video below, I tried and failed to get it to start at Willie’s talk. It begins at 47:30, but you’re also encouraged to watch the whole symposium.
Willie Buchser on using all of your spatial transcriptomic imaging data (begins at 47:30)