Fig. 1

scRNA-seq cell annotation and doublet inference a. Expression patterns of selected cell type-specific marker genes across cell types. b. Annotated cell types in the UMAP clusters (a) revealing 9 distinct cell types. Cell types were annotated using automatic annotation tools (SingleR and ScType) and manual refinement based on canonical cell type-specific markers (shown in 1a) c. Cell proportions in normal and tumour tissues. Normal tissues are dominated by fibroblasts and stromal cells, while tumour tissues contain more macrophages and a large number of tumour cells d. Proportion of normal and/or tumour tissues by clinical outcomes. Only complete remission samples contain cells from normal tissues. Partial remission and progressive disease samples consist of tumour tissues only. e. Cell distributions/proportions in tumour tissues from different clinical outcomes. The cancer cell fraction increases, while the immune cell fraction decreases as clinical outcomes worsen. f–h. Physical cell interaction network obtained from doublets in patients with (f) progressive diseases, (g) complete remission, and (h) partial remission generated using the Neighbor-seq method. Lines are coloured by counts; red to green representing low to high counts respectively. i. Heatmap comparing physical cell interaction types by clinical outcome. The physical interaction (doublet) counts were normalised by the number of cells in each clinical outcome and scaled by a factor of 10,000. CR, complete remission; PD, progressive diseases; PR, partial remission j Distribution of cancer-stromal doublets by tissue sites. PR and PD doublets were exclusively from the peritoneum whereas CR doublets came largely from the peritoneum but also arose from the omentum and ovaries. This was because samples for the PR and PD groups were taken only from the peritoneum but the CR samples were taken from the peritoneum. omentum and ovaries.