Project Summary / Abstract
Upcoming brain-wide descriptions of synaptic connectivity are poised to transform our understanding of brain
circuitry in the same way single-cell genomics has revolutionized our understanding of cell type diversity. The
challenge of relating whole-brain wiring diagrams to cell-type genetic properties must be overcome in order to
fully realize the potential of these datasets. Very few techniques generate multi-modality, "Rosetta stone"
datasets needed to link cell types to connectivity, and none presently have the throughput to do so across an
entire mammalian brain. In this proposal, we address key limitations that currently prevent such techniques from
scaling to meet the throughput of whole-mouse-brain connectivity initiatives, and develop the computational
frameworks needed to bind cell types to wiring diagrams.
The Patch-seq method links the full gene expression profile of single neurons with their fundamental properties,
including local morphology and electrophysiology1,2. In Aim 1, we will automate the Patch-seq technique to allow
parallelization and scaling sufficient for whole mouse brain coverage. This will be achieved by integrating and
optimizing recently developed methods for patch clamp automation, including pipette cleaning, cell detection,
and machine learning approaches to cell identification and tracking. Developments will be fully documented and
packaged for dissemination to lower barriers to access and further improve throughput via collaborative data
generation.
Similarly, methods for reconstructing the brain-wide full morphology of single neurons provides simultaneous
access to their local morphology and long-range projection targets. In Aim 2, we will improve and extend the
quality, efficiency, and capability of our automatic morphological reconstruction pipeline by adopting new
approaches to reconstruction (e.g., a hierarchy of deep learners), and testing advanced methods for tissue
processing and imaging across our Patch-seq and Full Morphology data generation pipelines. Automated
reconstruction methods will be trained and tested on gold standard data. Tools and data will be collaboratively
generated and publicly shared.
In Aim 3, we will develop new computational frameworks to link whole-brain connectivity datasets to multi-
modality cell type datasets. Powered by the throughput achieved in Aims 1 and 2, we will develop, apply, and
share machine learning-based data analysis methods to synthesize the observations collected from individual
platforms to achieve an integrated and predictive understanding of neuronal identity. This approach, which
facilitates cell type assignment, cross-modality integration and inference, and characterization of the
discreteness and continuity of fundamental cellular properties within and across types, will be scaled to achieve
whole mouse brain coverage.
