Notebooks
Categories
Cells
Premium
BioTuring
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
BioTuring
scVI-tools (single-cell variational inference tools) is a package for end-to-end analysis of single-cell omics data primarily developed and maintained by the Yosef Lab at UC Berkeley. scvi-tools has two components
- Interface for easy use of a range of probabilistic models for single-cell omics (e.g., scVI, scANVI, totalVI).
- Tools to build new probabilistic models, which are powered by PyTorch, PyTorch Lightning, and Pyro.
BioTuring
Single-cell RNA sequencing (scRNA-seq) data often encountered technical artifacts called "doublets" which are two cells that are sequenced under the same cellular barcode.
Doublets formed from different cell types or states are called heterotypic and homotypic otherwise. These factors constrain cell throughput and may result in misleading biological interpretations.
DoubletFinder (McGinnis, Murrow, and Gartner 2019) is one of the methods proposed for doublet detection. In this notebook, we will illustrate an example workflow of DoubletFinder. We use a 10x Genomics dataset which captures peripheral blood mononuclear cells (PBMCs) from a healthy donor stained with a panel of 31 TotalSeq™-B antibodies (BioLegend).
BioTuring
Single-cell RNA sequencing (scRNA-seq) protocols often face challenges in measuring the expression of all genes within a cell due to various factors, such as technical noise, the sensitivity of scRNA-seq techniques, or sample quality. This limitation gives rise to a need for the prediction of unmeasured gene expression values (also known as dropout imputation) from scRNA-seq data.
ADImpute (Leote A, 2023) is an R package combining several dropout imputation methods, including two existing methods (DrImpute, SAVER), two novel implementations: Network, a gene regulatory network-based approach using gene-gene relationships learned from external data, and Baseline, a method corresponding to a sample-wide average..
This notebook is to illustrate an example workflow of ADImpute on sample datasets loaded from the package. The notebook content is inspired from ADImpute's vignette and modified to demonstrate how the tool works on BioTuring's platform.
Trends
BioTuring
An R toolkit for accurate and efficient estimation of cell composition ('decomposition') from bulk expression data with single-cell information.
Bisque provides two modes of operation:
* Reference-based decomposition: This method utilizes single-cell data to decompose bulk expression. Bisque assumes that both single-cell and bulk counts are measured from the same tissue. Specifically, the cell composition of the labeled single-cell data should match the expected physiological composition. While Bisque doesn't explicitly require matched samples, Bisque expect having samples with both single-cell and bulk expression measured will provide more accurate results.
* Marker-based decomposition: This method utilizes marker genes alone to decompose bulk expression when a reference profile is not available. Single-cell data is not explicitly required but can be used to identify these marker genes. This method captures relative abundances of a cell type across individuals. Note that these abundances are not proportions, so they cannot be compared between different cell types.