Assessing GPT-4 for cell type annotation in single-cell RNA-seq analysis.

Here we demonstrate that the large language model GPT-4 can accurately annotate cell types using marker gene information in single-cell RNA sequencing analysis. When evaluated across hundreds of tissue and cell types, GPT-4 generates cell type annotations exhibiting strong concordance with manual annotations. This capability can considerably reduce the effort and expertise required for cell type annotation. Additionally, we have developed an R software package GPTCelltype for GPT-4's automated cell type annotation.

A systematic evaluation of computational methods for cell segmentation.

Cell segmentation is a fundamental task in analyzing biomedical images. Many computational methods have been developed for cell segmentation and instance segmentation, but their performances are not well understood in various scenarios. We systematically evaluated the performance of 18 segmentation methods to perform cell nuclei and whole cell segmentation using light microscopy and fluorescence staining images. We found that general-purpose methods incorporating the attention mechanism exhibit the best overall performance. We identified various factors influencing segmentation performances, including image channels, choice of training data, and cell morphology, and evaluated the generalizability of methods across image modalities. We also provide guidelines for choosing the optimal segmentation methods in various real application scenarios. We developed Seggal, an online resource for downloading segmentation models already pre-trained with various tissue and cell types, substantially reducing the time and effort for training cell segmentation models.

SifiNet: a robust and accurate method to identify feature gene sets and annotate cells.

SifiNet is a robust and accurate computational pipeline for identifying distinct gene sets, extracting and annotating cellular subpopulations, and elucidating intrinsic relationships among these subpopulations. Uniquely, SifiNet bypasses the cell clustering stage, commonly integrated into other cellular annotation pipelines, thereby circumventing potential inaccuracies in clustering that may compromise subsequent analyses. Consequently, SifiNet has demonstrated superior performance in multiple experimental datasets compared with other state-of-the-art methods. SifiNet can analyze both single-cell RNA and ATAC sequencing data, thereby rendering comprehensive multi-omic cellular profiles. It is conveniently available as an open-source R package.

PreTSA: computationally efficient modeling of temporal and spatial gene expression patterns.

Modeling temporal and spatial gene expression patterns in large-scale single-cell and spatial transcriptomics data is a computationally intensive task. We present PreTSA, a method that offers computational efficiency in modeling these patterns and is applicable to single-cell and spatial transcriptomics data comprising millions of cells. PreTSA consistently matches the results of state-of-the-art methods while significantly reducing computational time. PreTSA provides a unique solution for studying gene expression patterns in extremely large datasets.

GPT-4V exhibits human-like performance in biomedical image classification.

We demonstrate that GPT-4V(ision), a large multimodal model, exhibits strong one-shot learning ability, generalizability, and natural language interpretability in various biomedical image classification tasks, including classifying cell types, tissues, cell states, and disease status. Such features resemble human-like performance and distinguish GPT-4V from conventional image classification methods, which typically require large cohorts of training data and lack interpretability.

A systematic evaluation of computation methods for cell segmentation.

Cell segmentation is a fundamental task in analyzing biomedical images. Many computational methods have been developed for cell segmentation, but their performances are not well understood in various scenarios. We systematically evaluated the performance of 18 segmentation methods to perform cell nuclei and whole cell segmentation using light microscopy and fluorescence staining images. We found that general-purpose methods incorporating the attention mechanism exhibit the best overall performance. We identified various factors influencing segmentation performances, including training data and cell morphology, and evaluated the generalizability of methods across image modalities. We also provide guidelines for choosing the optimal segmentation methods in various real application scenarios. We developed Seggal, an online resource for downloading segmentation models already pre-trained with various tissue and cell types, which substantially reduces the time and effort for training cell segmentation models.

Spatial transcriptomics reveals segregation of tumor cell states in glioblastoma and marked immunosuppression within the perinecrotic niche.

Glioblastoma (GBM) remains an untreatable malignant tumor with poor patient outcomes, characterized by palisading necrosis and microvascular proliferation. While single-cell technology made it possible to characterize different lineage of glioma cells into neural progenitor-like (NPC-like), oligodendrocyte-progenitor-like (OPC-like), astrocyte-like (AC-like) and mesenchymal like (MES-like) states, it does not capture the spatial localization of these tumor cell states. Spatial transcriptomics empowers the study of the spatial organization of different cell types and tumor cell states and allows for the selection of regions of interest to investigate region-specific and cell-type-specific pathways. Here, we obtained paired 10x Chromium single-nuclei RNA-sequencing (snRNA-seq) and 10x Visium spatial transcriptomics data from three GBM patients to interrogate the GBM microenvironment. Integration of the snRNA-seq and spatial transcriptomics data reveals patterns of segregation of tumor cell states. For instance, OPC-like tumor and NPC-like tumor significantly segregate in two of the three samples. Our differentially expressed gene and pathway analyses uncovered significant pathways in functionally relevant niches. Specifically, perinecrotic regions were more immunosuppressive than the endogenous GBM microenvironment, and perivascular regions were more pro-inflammatory. Our gradient analysis suggests that OPC-like tumor cells tend to reside in areas closer to the tumor vasculature compared to tumor necrosis, which may reflect increased oxygen requirements for OPC-like cells. In summary, we characterized the localization of cell types and tumor cell states, the gene expression patterns, and pathways in different niches within the GBM microenvironment. Our results provide further evidence of the segregation of tumor cell states and highlight the immunosuppressive nature of the necrotic and perinecrotic niches in GBM.

Differences in microenvironment of lung cancer and pleural effusions by single-cell RNA sequencing.

BACKGROUND: Direct comparison of tumor microenvironment of matched lung cancer biopsies and pleural effusions (PE) from the same patients is critical in understanding tumor biology but has not been performed. This is the first study to compare the lung cancer and PE microenvironment by single-cell RNA sequencing (scRNA-seq). METHODS: Matched lung cancer biopsies and PE were obtained prospectively from ten patients. We isolated CD45+ cells and performed scRNA-seq to compare the biopsies and PE. RESULTS: PE had a higher proportion of CD4+ T cells but lower proportion of CD8+ T cells (False detection rate, FDR = 0.0003) compared to biopsies. There was a higher proportion of naïve CD4+ T cells (FDR = 0.04) and naïve CD8+ T cells (FDR = 0.0008) in PE vs. biopsies. On the other hand, there was a higher proportion of Tregs (FDR = 0.04), effector CD8+ (FDR = 0.006), and exhausted CD8+ T cells (FDR = 0.01) in biopsies. The expression of inflammatory genes in T cells was increased in biopsies vs. PE, including TNF, IFN-É£, IL-1R1, IL-1R2, IL-2, IL-12RB2, IL-18R1, and IL-18RAP (FDR = 0.009, 0.013, 0.029, 0.043, 0.009, 0.013, 0.004, and 0.003, respectively). The gene expression of exhaustion markers in T cells was also increased in tumor biopsies including PDCD1, CTLA4, LAG 3, HAVCR2, TIGIT, and CD160 (FDR = 0.008, 0.003, 0.002, 0.011, 0.006, and 0.049, respectively). CONCLUSIONS: There is a higher proportion of naïve T cells and lower proportion of exhausted T cells and Tregs in PE compared to lung cancer biopsies, which can be leveraged for prognostic and therapeutic applications.

A machine learning algorithm improves the diagnostic accuracy of the histologic component of antibody mediated rejection (AMR-H) in cardiac transplant endomyocardial biopsies.

BACKGROUND: Pathologic antibody mediated rejection (pAMR) remains a major driver of graft failure in cardiac transplant patients. The endomyocardial biopsy remains the primary diagnostic tool but presents with challenges, particularly in distinguishing the histologic component (pAMR-H) defined by 1) intravascular macrophage accumulation in capillaries and 2) activated endothelial cells that expand the cytoplasm to narrow or occlude the vascular lumen. Frequently, pAMR-H is difficult to distinguish from acute cellular rejection (ACR) and healing injury. With the advent of digital slide scanning and advances in machine deep learning, artificial intelligence technology is widely under investigation in the areas of oncologic pathology, but in its infancy in transplant pathology. For the first time, we determined if a machine learning algorithm could distinguish pAMR-H from normal myocardium, healing injury and ACR. MATERIALS AND METHODS: A total of 4,212 annotations (1,053 regions of normal, 1,053 pAMR-H, 1,053 healing injury and 1,053 ACR) were completed from 300 hematoxylin and eosin slides scanned using a Leica Aperio GT450 digital whole slide scanner at 40X magnification. All regions of pAMR-H were annotated from patients confirmed with a previous diagnosis of pAMR2 (>50% positive C4d immunofluorescence and/or >10% CD68 positive intravascular macrophages). Annotations were imported into a Python 3.7 development environment using the OpenSlide™ package and a convolutional neural network approach utilizing transfer learning was performed. RESULTS: The machine learning algorithm showed 98% overall validation accuracy and pAMR-H was correctly distinguished from specific categories with the following accuracies: normal myocardium (99.2%), healing injury (99.5%) and ACR (99.5%). CONCLUSION: Our novel deep learning algorithm can reach acceptable, and possibly surpass, performance of current diagnostic standards of identifying pAMR-H. Such a tool may serve as an adjunct diagnostic aid for improving the pathologist's accuracy and reproducibility, especially in difficult cases with high inter-observer variability. This is one of the first studies that provides evidence that an artificial intelligence machine learning algorithm can be trained and validated to diagnose pAMR-H in cardiac transplant patients. Ongoing studies include multi-institutional verification testing to ensure generalizability.