Tapping into the Rhythms That Lead to Predictions in the Brain

CNS 2025 Q&A with André Bastos

In some moments in time, technology seems to catch up with theory in powerful ways to elucidate new truths about fundamental processes in the brain. Now is that moment for understanding how brain rhythms coordinate to make everyday predictions that guide our learning and decision-making, says André Bastos, a cognitive neuroscientist who leads the Cognition, Computation, and Consciousness lab at Vanderbilt University. 

“We’ve really seen in the last few years a total revolution in our tools and our methods, and, increasingly, in our concepts and models,” says Bastos, who is a co-recipient of the CNS 2025 Young Investigator Award. His work, using a new method for high-density, multi-area recordings in the brain, is giving researchers a new window into the brain activity that leads to predictive processing – challenging seminal past work on predictive coding and creating a new model for moving forward.

I spoke with Bastos about the evolution of this work, including the new findings he will be presenting in Boston at CNS 2025, and what predictive processing really is, as well as how he got started in the field

CNS: How did you first become interested in cognitive neuroscience?

Bastos: I was totally fascinated by the idea of brain rhythms being this orchestra in the brain, conducting activity between one part of the brain with one type of wave and another part of the brain with another type of wave. And I was blown away by the idea that the integration of these activities may form the neural basis of consciousness. There was a 2001 paper by Francisco Varela that described that and really inspired me as an undergraduate at UC Berkeley studying cognitive science. 

CNS: How did you end up specifically studying predictive processing in the brain?

Bastos: I wanted to relate the work on brain rhythms to attention because I thought that would be the most rock-solid empirical evidence one could provide for the conductor metaphor actually using rhythms to guide cognition. That brought me to working with Ron Mangun and Marty Usrey at UC Davis to study visual and thalamic coordination and with Pascal Fries in the Donders Center for Cognitive Neuroimaging in the Netherlands to study ECoG [electrocorticography] in monkeys. It was a unique opportunity in my PhD where I was able to sort of do a “choose your own adventure” type of thing, as I had received an NSF Graduate Research Fellowship Award and a Fulbright award at the same time. I was then able to apply that work in Karl Friston’s lab at University College London to learn about predictive coding and its possible neurobiological implementation. It was with Karl that I worked on a paper about canonical microcircuits for predictive coding that forms the basis of my current work. 

After working mostly at the level of theory and seeing the power of computational-level thinking during my PhD, I realized I need to be able to learn how to do the empirical work myself, and that’s what brought me then to do my postdoc with Earl Miller at MIT. The combination of these very powerful concepts and theories and data, which I learned from Pascal, Karl, Marty, Ron and Nancy Kopell, still drives my work today on predictive routing. I am very indebted and thankful to my mentors.

CNS:  So what is predictive routing?

Bastos: Predictive routing is just the idea that you form predictions based on past experiences. These predictions make perception and cognition smoother, faster, and easier, compared to situations we have not previously encountered. The proposed neural implementation of that is that the predictions are associated with different rhythms in the brain. The rhythm most associated with prediction is the beta frequency range, oscillations around 20 hertz. The beta rhythm coordinates neurons and can form plastic, dynamic ensembles or groupings of neurons, and that is what learning is – forming new connections between neurons that didn’t exist before. 

We think of those beta ensembles as representing what our current predictions are for us right now, at this moment. So, for example, if you show me a new book or describe a new idea that I’ve never seen before, I can quickly make an association based on a combination of current information and past experience. These associations coalesce together to form new, complex, and dynamic predictions. Our work shows that this type of predictive coding is more prevalent in higher-order areas like the prefrontal cortex rather than sensory cortices. 

CNS: In your CNS 2025 talk to accept the Young Investigator Award, you will be presenting new data about how those beta waves interact with other waves during predictive routing. Can you give us a brief preview of that?

Bastos: We hypothesize that the beta waves inhibit the gamma, and we’re getting more and more causal evidence for that. The gamma is associated with processing of sensory data from the environment. So when you’re forming predictions, you are inhibiting the stuff that you don’t need to encode very strongly or pay attention to very strongly, because it’s already consistent with what you know.  When we talk about top-down and bottom-up processing, we are seeing gamma as the bottom-up sensory stream and beta as a control, sculpting the flow of sensory information. And that’s the basis for predictive routing, which builds off previous work by Earl Miller and others.

CNS: Your abstract for the talk mentions a new tool that’s helping with this work, Multi-Area, high-Density, Laminar Neurophysiology (MaDeLaNe). Can you explain how you are using this technology?

Bastos: MaDeLaNe puts us in a position we’ve never been in before, not by a long shot, to be able to record from a large amount of the brain at the same time. In the past, if you were lucky, you might get a handful of cells or perhaps be recording from two or three areas; this is in an animal model or through work with a human patient who is having electrodes surgically implemented for monitoring. Now what we’ve done is significantly advanced our capabilities, to be able to record hundreds of neurons across a particular area and at high density, with hundreds of channels packed within a few millimeters. So from 2014, when I began my postdoc, to now, we’ve shifted up an order of magnitude, looking at many more neurons at a time over many more areas. 

CNS: And so this is leading to some new ideas about how neurons work to create the predictive coding that drives predictive routing, right?

Bastos: Yes! We started with a theory of predictive coding that took us to developing a new technique in order to really test the theory. And as often happens in science, our perfect theory is coming up against the ugly facts of what we found, which is now pushing us toward a modification and nuancing of the theory. We now have evidence that what we thought was a ubiquitous code for predictions is actually a sparse code, with the main driver being neurons in higher-order areas utilizing beta rhythms (rather than sensory areas, which were previously hypothesized to compute prediction “errors”). And we now call that predictive routing. So we’re seeing one cycle of science, between theory and experiment, closing, now that we’re releasing these results out to the world, with the hope that it can improve our understanding of how predictions are created in the brain.

CNS: What are you most looking forward to at CNS 2025?

Bastos: Definitely seeing all my mentors who are in attendance! Being able to reflect now on this journey that we’ve all been on is going to be really special. 

CNS: Is there anything else you would like to add?

Bastos: My research is taxpayer funded through the NIH and NSF. That support is critical to science moving forward – and critical to our advancement of treatments for disorders, innovation in the economy, and our global competitiveness.

-Lisa M.P. Munoz