Single-cell transcriptomic landscape deciphers novel olfactory neuroblastoma subtypes and intratumoral heterogeneity

Abstract

Object: To explore the intra- and inter-tumoral heterogeneity of olfactory neuroblastoma (ONB). To discover ONB subtypes with distinctive expression features and different putative cellular origins and potential treatment strategies. To depict the tumor microenvironment (TME) landscape and cellular crosstalk with tumor cells. Methods: Single-cell RNA sequencing of 10 treatment-naive ONBs was conducted to explore the cell types and distinguish malignant cells, so as to form an innovative classification of tumors with diverse molecular markers and cellular origins, validated using bulk mRNA profiles both internally and externally. TME components of ONB was depicted by sub-clustering of major immune and stromal clusters, pathway enrichment analysis and cellular interaction analysis. Results: Cell clusters were classified into 6 immune lineages, 3 non-immune stromal lineages including and one epithelial cluster. Malignant cells determined by infer CNV reflected diverse cellular functions including olfactory neurodevelopment, cell cycling and mesenchyme features. A ternary diagram confirming a tripartite structure of cellular phenotypes across ONB malignant cells was established. Similarities between the three subtypes and diverse olfactory lineages were evaluated respectively. We explored molecules underlying malignant transition from normal OM to ONB. Subclusters of TME components of ONB and their cell-cell interactions with tumor cells were depicted. Conclusion: An innovative three-classification of ONB was established via scRNA analysis. Markers for categorizing tumor samples into new subtypes were elucidated. Different responses towards certain chemotherapy regimens could be inferred according to the molecular features of three tumor types. Relative abundance of immunosuppressive TAMs indicated the benefits of immunotherapies targeting myeloid cells.

FExGAN-Meta: Facial Expression Generation with Meta Humans

“…The second group of methods includes faster two-stage approaches [44,49,47,106,105,8,81,28,23] that start by generating multiple views of the object using a text-to-image or -video model [52,13] tuned to output multiple views of the object followed by per-scene optimization using NeRF [56] or 3DGS [37]. However, per-scene optimization requires several highly-consistent views which are difficult to generate reliably.…”