Skip to main content
Fig. 2 | Genome Medicine

Fig. 2

From: Transfer learning between preclinical models and human tumors identifies a conserved NK cell activation signature in anti-CTLA-4 responsive tumors

Fig. 2

CoGAPS identifies gene signatures related to immune cell lineage and treatment response in mouse intratumoral immune cell scRNA-seq data. A Overview of the pipeline to relate preclinical and clinical mechanisms of action of therapy using transfer learning. First, CoGAPS, a non-negative matrix factorization algorithm is applied to scRNA-seq data of ICI-treated mouse tumors. Matrix factorization algorithms are unsupervised learning methods that can distinguish low-dimensional gene and cell features (latent spaces) associated with therapeutic responses without prior knowledge of gene regulation or cell type classification. Next, the transfer learning method projectR, is used to project the transcriptional signatures representing the latent spaces (or patterns) identified by CoGAPS into an independent dataset of human tumors treated. Finally, the cell weights representing relative usage of each pattern in the new human dataset can be computationally assessed for relationships to clinical outcomes and as the basis to prioritize candidates for experimental validation. B UMAP-dimension reduction of droplet-based scRNA-seq of intratumoral immune cells from ICI-treated mouse sarcomas [4]. Samples are colored by annotated cell types (left) and by treatment (right). C Hierarchical clustered heatmap of 21 CoGAPS patterns demonstrating segregation by immune cell lineage. Rows are individual cells, with row annotations designating cell type. Columns represent different CoGAPS patterns. D UMAP-dimension reduction colored by CoGAPS pattern 13 weights illustrates a cell type specific signature within the macrophages/monocytes. E Boxplot of pattern 13 weights in individual macrophage/monocyte cells, faceted by treatment group. Pattern 13 is associated with cells treated with control monoclonal antibody. Significant differences in mean pattern 7 weight between treatment groups are indicated by asterisks where p-values < 0.05 = *, < 0.01 = **, and < 0.001 = ***. F UMAP-dimension reduction colored by CoGAPS pattern 12 weights illustrates a cell type specific signature within the macrophages/monocytes. G Boxplot of pattern 12 weights in individual macrophage/monocyte cells, faceted by treatment group. Pattern 12 is associated with cells treated with anti-PD-1. Significant differences in mean pattern 7 weight between treatment groups are indicated by asterisks where p-values < 0.05 = *, < 0.01 = **, and < 0.001 = ***

Back to article page