Revealing Spatial Gene Patterns and Interactions in Mouse Brain via Stability-Driven NMF

• Activity Number: 1 - Invited E-Poster Session
• Type: Invited
• Date/Time: Sunday, August 2, 2020 : 12:30 PM to 3:30 PM
• Sponsor: IMS
• Author(s): Yu Wang* and Reza Abbasi-Asl and Nathan Gouwens and Zizhen Yao and Bosiljka Tasic and Hongkui Zeng and Anton Arkhipov and Bin Yu
• Companies: UC Berkeley and UC San Francisco and Allen Institute and Allen Institute and Allen Institute and Allen Institute and Allen Institute and UC Berkeley
• Keywords: NMF; spatial gene expression; gene networks; PCA

Abstract

Identifying the functional regions in the brain from spatial gene expression profiles is one of the crucial tasks in computational biology. A mapping from genes to brain regions sheds light on the functional roles of genes as well as the relationship between marker genes in brain regions. Identifying such a mapping between genes and brain regions has been challenging to automate and manual labeling requires intensive labor. Here, we propose an automated unsupervised framework to partition the 3D gene expression profiles into brain regions. Our framework is based on stability-driven non-negative matrix factorization (staNMF) with sparsity constraints. Using Allen Brain Atlas, we show that our framework partitions the 3D gene expression profiles in the mouse brain into spatially segmented principle patterns (PPs). The obtained PPs correspond to the manually segmented brain regions in the Atlas. Using bootstrapping techniques, we demonstrate that our framework provides more stable PPs compared to principal component analysis. Finally, we used the learned PPs to construct local transcription factor regulatory gene networks.