Human Brain Single-Cell Genomics Explorer

University Of California Santa Cruz, Santa Cruz CA

Investigators

Abstract

PROJECT SUMMARY / ABSTRACT There has been an explosion of genomics data in neuroscience that bridges insights from single-cell genomic data to brain function and diseases. Data on hundreds of millions of cells are being generated by consortia such as BICAN/BICCN, PyschENCODE, SEA-AD, SCORCH, and SSPsyGene. As many different cell taxonomies are being generated, the neuroscience field needs a single-cell brain resource for researchers to explore and use as a reference model for their datasets. A huge semi-manual effort would be not only nearly impossible given the data size, but also would quickly become obsolete. Only automated efforts will be able to address the need for a durable, flexible, and nimble neuroscience cell reference. We propose to use cutting-edge AI techniques to build a Universal Cell Embedding for Brain (UCE-Brain), a foundational brain cell taxonomy model across hundreds of millions of cells. We will develop the UCSC Human Brain Single-cell Genomics Explorer to enable users to access and interactively explore the UCE-Brain, the reference cell taxonomy, as well as map new data into the reference. The Data Explorer will be capable of interactive data visualization of the hundreds of millions of cells in a 2D embedding space; the Cell Curator will explore the BICAN taxonomy annotations as well as gathering community-driven cell types; the Cell Mapper will offer two methods to map new data to the reference models and visualize the results; and Tools and Documentation. We will incorporate a Large Language Model (LLM) into the Cell Curator and Cell Mapper to parse users’ requests, provide a natural language summary of the results in a chatbot-style interaction, and integrate user feedback. Our team’s diverse expertise and strong history of successful collaboration will ensure the success of the project. Aim 1. Construct foundation models for human brain cells We will develop UCE-Brain, a brain-specific foundation model that will create high quality universal representations of brain single cell data. We will build UCE-Brain clusters that are most consistent with BICAN taxonomy in gold-standard publications, while still allowing the clusters to evolve in response to new datasets. Aim 2. Mapping new data into the reference cell type taxonomy. We will offer two methods for users to map and annotate their data. One method will allow users to upload their data to a server to map to UCE-Brain. A second method will allow users with sensitive data to map to a stripped-down model on their laptops. Aim 3. Develop the Cell Curator to promote community annotation of the reference cell type taxonomy. The Cell Curator tool will enable users to inspect and curate the reference cell clusters. We will conduct efforts to foster the neuroscience research community’s contributions. Aim 4. Develop the UCSC Brain Explorer website and harmonize single-cell sequencing data. The UCSC Brain Explorer website serves as the gateway to all functionality and resources developed in this application. Data harmonization will occur on Terra, where we will process datasets using the BICAN/BICCN pipelines. We will integrate Google Analytics to track aggregated user interactions.