Going with the Flow: Mapping Information in the Human Brain

Q&A with Michael W. Cole

Increasingly, cognitive neuroscientists are focusing on computation to better understand how information is stored and moves through the human brain. For Michael Cole, this work has included computer science at Apple and behavioral science at Berkeley, with him ultimately creating a cognitive neuroscience lab at Rutgers University that is taking new integrated approaches to mapping the brain’s networks.

Recognizing Cole’s already tremendous contributions to the field, CNS has awarded him one of the Young Investigator Awards for 2019. At the upcoming CNS annual meeting in San Francisco Cole will highlight work on activity flow — the movement of neural activity over neural connections throughout the brain. “For instance, when stopping at a red light while driving, activity enters your visual system and flows through a complex cascade to eventually cause activity in your motor cortex that moves your foot to the brake,” he explains. “These activity flow events provide a clear link between activity and connectivity, as well as with cognition and behavior.”

Cole spoke with CNS about this work, its implications for the field, and what he most looks forward to at CNS 2019 in San Francisco.

CNS: How did you personally become interested in studying cognitive neuroscience?

Cole: I was undecided as an undergraduate at UC Berkeley between majoring in computer science, psychology, or philosophy. I was ecstatic to come across a major that let me not have to decide: cognitive science. While fulfilling my cognitive science requirements, I took some neuroscience courses and it seemed clear to me that answers to many of the questions asked in the other cognitive science subfields would eventually be answered by neuroscience.

Even then I hedged my bet, working part-time at Apple (to perhaps make use of my new computer science skills) while volunteering in a cognitive neuroscience lab. I soon realized I had a deep passion for cognitive neuroscience. It has the right balance of being extremely important and being largely undiscovered. This suggests there are some exciting new discoveries left to make in this field that will fundamentally improve the human condition.

CNS: Why do you study brain network organization specifically?

Cole: I study brain network organization in particular because I’ve been obsessed with computational modeling since I was an undergrad — but I keep worrying so much that any model I create won’t match reality that I’ve focused mainly on empirical neuroscience. It turns out that all computational models of the brain primarily compute cognition and behavior via connectivity patterns, hence the common term “connectionism” for artificial neural network modeling. So naturally, I’ve been wanting to focus on brain network organization as it is very likely the primary computational basis of cognition.

CNS: How is your lab going about mapping and understanding the human brain’s large-scale functional network organization?

Cole: We are standing on the shoulders of giants, building on a variety of methods that are individually-limited but complementary in their usefulness when combined to map out connectivity patterns in the human brain. We’re primarily using functional MRI (fMRI), electroencephalography (EEG), and magnetoencephalography (MEG). These methods allow us to record brain activity while participants rest and perform a variety of tasks, and then we look at the similarity of the ongoing brain activity to infer likely patterns of interaction between different parts of the brain.

I realized activity flow processes are used in both artificial neural networks and real biology, making them a linking principle across areas of neuroscience and engineering as well as across activity and connectivity measures.

CNS: How does your approach differ from others?

Cole: Our approach differs from past efforts in several respects. First, we join several groups in using the tools of network science to understand brain network organization. Network science is a cross-disciplinary field that develops tools to identify things like hubs – highly interconnected nodes in a network, akin to extremely busy airports in the world air transportation network. We have also used community detection algorithms, which identify clusters of highly interconnected nodes. We are also one of relatively few groups to be using dynamic network neuroscience methods, primarily with task-state functional connectivity (see Cole et al. 2014 and Cole et al. 2019).

Second, we are working hard to integrate this new connectivity-oriented perspective that is becoming increasingly popular in cognitive neuroscience with the perennially popular task-evoked activation perspective. We are seeing a tendency to throw the baby out with the bathwater when it comes to moving on from localizing functions based on task-evoked activations to mapping the brain’s connectivity. Instead, we should be leveraging the strengths of each of these approaches to develop an integrated understanding of brain function.

Along these lines, we have been developing several approaches under the realization that connectivity is highly complementary to task-evoked activity for understanding brain function. This reflects the well-established fact that brain activity flows over brain connections to determine activity in other parts of the brain.

We are a long way from being able to measure these activity flows in all of their detail. This is the case for human neuroimaging primarily because of limited spatiotemporal resolution, while this is true in animal models because of limited spatial coverage with current recording methods. However, by being clear that our goal is to estimate and understand these activity flows, we have been able to make some progress. This insight came for me in late-2013/early-2014 as I realized activity flow processes are used in both artificial neural networks and real biology, making them a linking principle across areas of neuroscience and engineering as well as across activity and connectivity measures.

CNS: Can you give an example of a recent finding that is coming out of research in this area that you’ll be presenting on at the meeting?

Cole: First, we recently used the activity flow principle with fMRI to see if resting-state functional connectivity could be used to “transform” healthy task activity patterns into pre-Alzheimer’s disease task activity patterns. The key idea is that changes in functional connectivity might underlie, at least in part, the cognitive difficulties in Alzheimer’s disease by distorting task-evoked activity flow processes. This hypothesis was supported by the finding that we can predict abnormal task-evoked activations via activity flow mapping. One thing that’s fairly surprising to me is that we can accurately predict abnormal task activations without needing the participant to actually perform the task; we only need the participant’s resting-state fMRI data to parameterize the activity flow simulation.

We have several other recently findings I’ll be going into as well, along with current limitations with the activity flow framework we are also making progress on improving.

CNS: Have any of the recent results from your lab been surprising? If so, how so?

Cole: This entire line of research has been surprising to me.

When I developed the activity flow mapping approach, I was actually trying to show that this wouldn’t work well, at least with fMRI resting-state functional connectivity. I really didn’t expect fMRI to be the kind of brain activity measure to be able to show activity flow processes well. And I especially didn’t expect resting-state functional connectivity to be the kind of connectivity to be able to show activity flow processes well.

There are several remaining limitations to what activity flow mapping with fMRI can do, of course. See the original activity flow mapping paper for more info on this. We are working on multiple fronts to further improve the activity flow framework, and we invite others to join us in this effort.

CNS: How do you think the fundamental work you are doing will ultimately translate to other areas or clinical work?

Cole: As mentioned earlier, we have already begun applying activity flow mapping to clinical data, in the form of using resting-state fMRI data to predict abnormal task-evoked brain activations in pre-Alzheimer’s disease. This is helping to establish the functional role of resting-state functional connectivity alterations in cognitive deficits. We expect this approach to be useful for a variety of other clinical conditions as well.

For instance, we recently found that there are only minor, but likely very important, alterations in resting-state functional connectivity for individuals with schizophrenia, autism spectrum disorder, or attention-deficit/hyperactivity disorder. We hypothesize that these small connectivity alterations relative to healthy individuals likely nonetheless have large effects on task brain activations, given the activity flow principle combined with how widespread each region’s connectivity tends to be. We are excited to test this hypothesis in the near future.

CNS: What do you most want people to understand about your work?

Cole: I want people to understand that we very much want to help others build on what we’ve done so far. There is so much that can be done with the tools we have released, and we genuinely believe that the activity flow framework, especially as we and others improve its methodology, can lead us to a more unified understanding of brain function. This is because it applies a core unifying theoretical construct in neuroscience to link disparate forms of data in a principled manner.

We need the full confluence of fields contributing to cognitive neuroscience to maximize the creativity and eventual success of cognitive neuroscientists, and I’m excited to see all of this coming together at CNS 2019.

CNS: What are the next steps for your work?

Cole: Having the ambitious goal of estimating the human brain’s large-scale activity flow processes underlying cognitive task performance has put a lot of the strengths and limitations of modern cognitive neuroscience into perspective for us. For instance, we recently found evidence that the way we and most others in the field have been looking at task-state functional connectivity has been wrong, and we found a solution that works well. Continuing this kind of logic, we are seeing that non-linearities in the relationship between neural populations are likely fairly important for activity flow computations. So, we are starting to look into non-linear functional connectivity measures as well.

Another major theme going forward is a realization that causal inference is central to virtually all aspects of cognitive neuroscience. This suggests recently-developed causal inference tools can be used to improve cognitive neuroscience. I’ll actually be giving a separate talk at CNS 2019 on this topic.

CNS: What are you most looking forward to about the CNS meeting in San Francisco?

Cole: Looking over the schedule, I’m seeing a long list of great talks, and I have no doubt the poster sessions will be lively and informative as well. I’ve been happy to see the full richness of cognitive neuroscience – touching on computation, clinical disorders, psychology, and animal neuroscience, to name a few contributing areas – becoming better represented at CNS recently. We need the full confluence of fields contributing to cognitive neuroscience to maximize the creativity and eventual success of cognitive neuroscientists, and I’m excited to see all of this coming together at CNS 2019.

CNS: Anything I didn’t ask you about that you’d like to add?

Cole: I’d like to add that I’m honored to be receiving this award, and I’m very grateful to my mentors and mentees for helping me get here. As a short sample of a long list, I’m grateful to my primary postdoc advisor Todd Braver and graduate advisor Walt Schneider for everything they taught me. And of course my mentees have been amazingly helpful and supportive, such as Taku Ito, Doug Schultz, Carrisa Cocuzza, Richard Chen, Marjolein Spronk, and Ravi Mill. There are of course others (in addition to friends and family) whose help has been important for my success, such as: Deanna Barch, Steve Petersen, Julie Fiez, Patryk Laurent, Alan Anticevic, Grega Repovs, Danielle Bassett, Mark D’Esposito, Steve Hanson, Denis Pare, and Bart Krekelberg. I couldn’t have done it without them!

-Lisa M.P. Munoz