Hava T. Siegelmann and Patrick Taylor
London E1W 1YW, UK
Portland, ME 04101
2nd floor
11th floor
Boston, MA 02115
2nd floor
London E1W 1LP, UK
Talk recording
We introduce three projects describing networks in the brain, the first two at a macro-scale in humans with the finest resolution at about 1mm, and the third derived from single-neuron level wet data in rodents. The first two research projects utilize resting-state fMRI and DTI to infer and build a network structure. In the first, we focus on a feed-forward interconnections that build up from sensory cortical areas, revealing an architectural principle that illustrates higher symbolic abstraction progressing from tangible input. The second and third projects emphasize recurrent structure and aim to provide, (2.1) a simulation that can predict fMRI sequences, and finally (2.2) compare network properties between the single-neuron level network dynamics in models of rat brain, versus the macro-level connectivity in the human brain.
Project 1: Though widely hypothesized, limited evidence exists that human brain functions organize in global gradients of abstraction. Our methods developed directed network models, by including input locations, the primary sensory cortices. The directional network confirmed previous models: regions related to awareness exist most connected to inputs, with higher-executive frontal regions least connected to inputs. The model was used in this work to guide analyses of fMRI databases (about 17,000 experiments or 1/4 of all existing fMRI literature). We tested whether a region's network depth predicted localization of abstract versus concrete functions from fMRI data. We found that it is not the depth per se, but a new measure (defined as depth-slope) we used to approximate an activation cloud's network depth, predicting abstraction. We objectively sorted stratified landscapes of cognition based on the slope measure, starting from grouped sensory inputs in parallel, progressing deeper into cortex. This exposed escalating amalgamation of function or abstraction with increasing network-depth slope, globally. Nearly 500 new participants confirmed our results. In conclusion, data-driven analyses defined a hierarchically ordered connectome, revealing a related continuum of cognitive function. Progressive functional abstraction over network depth may be a fundamental feature of brains, and is observed in artificial networks. Results pioneer statistically-directional macro-cortical network models.
Project 2.1 (human-macro): Macro-level whole brain network dynamics between regions can be modeled by a network of smaller regions, themselves compose of many fewer neurons. For example, a human brain with 86 billion neurons, clustered into 100 regions would have roughly 860 million neurons per region. Our model with 86,000 neurons in the whole brain has inter-region connectivity stochastically initialized proportional to the known inter-region connectivity between human brain regions (via DTI or rsfMRI). We study the self-similarity feature of this network. The model is built to predict fMRI experiments performed on two groups of patients, each following a ketogenic versus glucogenic diet, and is our BRAIN initiative NSF project
Project 2.2 (rodent-micro): We also initialize a version of the human-brain model, but with data based on reconstructed microcircuitry of the cortex of rats. The aim is to compare structure and dynamics among the two species and levels.