Figure 6. Regulon-based simulations from bulk RNA deep sequencing data suggest that mitral cell types have distinct projection targets.

(A) Schematic representation of strategy to integrate bulk RNA-seq and single nucleus RNA-seq data. (A1) Simulations use regulons inferred from sn-R1/R2/R3 data. A regulon consists of a transcription factor (TF) and all target genes (A, B, etc) that are activated by that transcription factor. (A2) When simulating nuclei from bulk RNA-seq data, the probability of selecting a given regulon is determined by the abundance of its transcription factor in the bulk RNA-seq dataset. (A3) Nuclei are simulated from each bulk dataset (AON-sim and PCx-sim) through random sampling of regulons with replacement. This method maintains broad differences between datasets while accounting for heterogeneity within each dataset. (B) AON-sim, PCx-sim and sn-R1/R2/R3 nuclei projected into a shared UMAP representation. Step 1: Blue indicates AON-sim nuclei and purple indicates PCx-sim nuclei. Step 2: Darker color indicates sn-R1/R2/R3 projection neurons. Step 3: Blue indicates AON-projecting simulated projection neurons (AON-simPN) and purple indicates pPCx-projecting simulated projection neurons (PCx-simPN). (C) Linear Discriminant Analysis (LDA) classifiers were trained on AON-simPN and PCx-simPN, then used to predict the projection target of sn-R1/R2/R3 mitral cells to investigate projection targets of sn-R1/R2/R3 derived types. Each row represents one mitral cell. Each column represents one of 1000 LDA classifiers. Blue indicates that the mitral cell was classified as AON-simPN, and purple indicates that the mitral cell was classified as PCx-simPN. Within each mitral cell type, cells are sorted vertically by their predicted projection target. (D) UMAP representation color-coded by predicted projection target. Cells in blue were predicted to be AON-simPN by all 1000 classifiers. Cells in purple were predicted to be PCx-simPN by all 1000 classifiers.