Seung Lab

Computational Neuroscience@Princeton

The first neuronal wiring diagram for a brain was recently released, and has leaped orders of magnitude beyond the 1986 wiring diagram of the C. elegans nervous system. This historic milestone showcases the revolutionary progress achieved in EM image acquisition and analysis over the past decade. The connectomic era has finally begun.

The Seung lab first applied convolutional nets to EM image analysis in 2007, and continued development of this approach resulted in the fly connectome and a wiring diagram of a cortical column. Now we are passing the baton to Zetta AI, which will continue innovating to create the artificial intelligence required to scale up connectomics to a mouse brain and beyond.

The Seung lab has shut down its EM image analysis effort, and refocused on making discoveries about brain function, as well as brain development and plasticity.

Interpretation of connectomes

We are devising concepts and methods for interpreting neuronal wiring diagrams. The fly connectome includes as a corollary the first wiring diagram of a visual system. By studying this wiring diagram, we have discovered a new circuit for form vision in the fly brain. We have also discovered that many inhibitory interneuron types function as a highly diverse set of normalization mechanisms in fly vision. Both of these discoveries draw on the striking analogy between the fly visual system and convolutional nets.

Reconstructing neural circuits

In ongoing collaborations, we are applying and refining connectomic technologies to reconstruct more fly connectomes (Mala Murthy), as well as a patch of mouse retina (Thomas Euler). We are reconstructing mouse neural circuits for memory (David Tank), decision making (Adrian Wanner and Jeff Lichtman), and reinforcement learning (Ilana Witten). These collaborations make use of the high throughput EM facility at the Princeton Neuroscience Institute. In many of the projects, neural circuit reconstruction is preceded by calcium imaging of neural activity in vivo.

Scaling up to mammalian brains

Today’s connectomic technologies are sufficient for reconstructing an entire fly brain, and are also being applied to millimeter-scale chunks of mammalian brains. A mouse brain is 1000× larger, and a human brain 1000× larger still. There is plenty of room at the top. We are participating in a “transformative project” of the NIH BRAIN Initiative that aims to scale up connectomics to a whole mouse brain. The Princeton Neuroscience Institute is the only site in the world with both of the EM image acquisition technologies that are being scaled up to the mouse connectome, beam-deflection transmission electron microscopy and multi-beam scanning electron microscopy.

Scaling down to molecules

The fly connectome was reconstructed from EM images with 4×4×40 nm3 voxels, which is sufficient for detecting chemical synapses and tracing the “wires” of the brain. This resolution might seem very fine, but is actually coarse compared to the 0.1 nm theoretical limit of EM. There is plenty of room at the bottom. Serial section EM tomography can improve resolution; the challenge is to deliver this improvement over much larger volumes than before. One can imagine, for example, imaging an entire fly brain at 4×4×4 nm3 or 2×2×2 nm3 resolution. This would reveal brain cell biology in fantastic detail, within the full context of neurons and their connections.