Logo: University of Southern California

Viterbi Team Learning How 100,000 Cells Collaborate to Remember

Ted Berger's lab creates a large-scale computational model of brain memory encoding
By: Eric Mankin
February 04, 2013 —


Ted Berger
Scientists at the USC Viterbi School of Engineering have created a one-tenth scale working computer model of a specific anatomical structure of a rat brain. The next step: full scale.

The brain section adjoins the hippocampus, a part of both rat and human brains that is essential to learning. Professor Ted Berger of the Viterbi School's Department of Biomedical Engineering has been studying hippocampal function for more than 30 years, attempting to understand how interconnections between large numbers of neurons create the behavioral effects seen in brain function.

New long-term memories are created by the connection of a part of the hippocampus called the entorhinal cortex (EC) with another part of the hippocampus called the dentate gyrus (DG). This connection, called the perforant path, mediates the selective activation of certain of some of the 1 million dentate gyrus cells in living rats. The electronic computer functional model created by the Berger group models 100,000 dentate gyrus cells stimulated by 12,000 entorhinal cortical cells.

The critical parameter is the pattern of intercommunication. Cells connect to certain other cells in complex degrees of separation patterns, directly connecting to some, while connecting indirectly or not at all with others.

To model this pattern, the Berger group simplified a detailed cell model it had previously developed for DG cells, which modeled the activity of individual cells well but made it difficult to correctly place them into their unique roles in the message flow.

Instead, the group modeled the branches of the cell connections tree using a computer program by Giorgio Ascoli called L-Neuron incorporating an algorithm created by computer scientist Dean Hillman.

The experiment performed on this computer model system was to energize the DG cell corpus with stimulation, and then observe the way the 100,000-cell model responded.

The response, according to a recent paper, was encouraging. The excitation pattern spread in way very consistent with what is seen in records of parallel biological mouse EC-DG gyrus stimulation, and the behavior seem to correctly reproduce critical patterns of feedback inhibition, variable spike timing, and changes in synaptic activity.

Next step: scale up. Full brain model? Not for a while. The lead author of the study was graduate student Phillip Hendrickson. Besides Berger, the other authors were two other biomedical engineering graduate students, Gene Yu and Brian Robinson.The work was funded by the ERC and DARPA.