E-spatial

Geneformer is a foundation transformer model pretrained on a large-scale corpus of ~30 million single cell transcriptomes to enable context-aware predictions in settings with limited data in network biology. Here, we will demonstrate a basic workflow to work with ***Geneformer*** models. These notebooks include the instruction to: 1. Prepare input datasets 2. Finetune Geneformer model to perform specific task 3. Using finetuning models for cell classification and gene classification application

In the realm of transcriptional dynamics, understanding the intricate interplay of regulatory proteins is crucial for deciphering processes ranging from normal development to disease progression. However, traditional RNA velocity methods often overlook the underlying regulatory drivers of gene expression changes over time. This gap in knowledge hinders our ability to unravel the mechanistic intricacies of these dynamic processes. scKINETICs (Key regulatory Interaction NETwork for Inferring Cell Speed) (Burdziak et al, 2023) offers a dynamic model for gene expression changes that simultaneously learns per-cell transcriptional velocities and a governing gene regulatory network. By employing an expectation-maximization approach, scKINETICS quantifies the impact of each regulatory element on its target genes, incorporating insights from epigenetic data, gene-gene coexpression patterns and constraints dictated by the phenotypic manifold.

The development of immune checkpoint-based immunotherapies has been a major advancement in the treatment of cancer, with a subset of patients exhibiting durable clinical responses. A predictive biomarker for immunotherapy response is the pre-existing T-cell infiltration in the tumor immune microenvironment (TIME). Bulk transcriptomics-based approaches can quantify the degree of T-cell infiltration using deconvolution methods and identify additional markers of inflamed/cold cancers at the bulk level. However, bulk techniques are unable to identify biomarkers of individual cell types. Although single-cell RNA sequencing (scRNAseq) assays are now being used to profile the TIME, to our knowledge there is no method of identifying patients with a T-cell inflamed TIME from scRNAseq data. Here, we describe a method, iBRIDGE, which integrates reference bulk RNAseq data with the malignant subset of scRNAseq datasets to identify patients with a T-cell inflamed TIME. Utilizing two datasets with matched bulk data, we show iBRIDGE results correlated highly with bulk assessments (0.85 and 0.9 correlation coefficients). Using iBRIDGE, we identified markers of inflamed phenotypes in malignant cells, myeloid cells, and fibroblasts, establishing type I and type II interferon pathways as dominant signals, especially in malignant and myeloid cells, and finding the TGFβ-driven mesenchymal phenotype not only in fibroblasts but also in malignant cells. Besides relative classification, per-patient average iBRIDGE scores and independent RNAScope quantifications were utilized for threshold-based absolute classification. Moreover, iBRIDGE can be applied to in vitro grown cancer cell lines and can identify the cell lines that are adapted from inflamed/cold patient tumors.

Single-cell RNA-seq datasets in diverse biological and clinical conditions provide great opportunities for the full transcriptional characterization of cell types. However, the integration of these datasets is challeging as they remain biological and techinical differences. **Harmony** is an algorithm allowing fast, sensitive and accurate single-cell data integration.