How GPT-4 Provides High Accuracy Cell-Type Annotations in Single Cell RNA Sequencing Experiments

How GPT-4 Provides High Accuracy Cell-Type Annotations in Single Cell RNA Sequencing Experiments

Table of Contents


Single-cell RNA sequencing (sc-RNAseq) is a technique to allow transcriptional profiling of complex biological systems at a cellular level. An essential part of sc-RNAseq data analysis is cell annotation. A new study by Hou W & Ji Z. (2024) demonstrates GPT-4, a powerful language model, to automate cell annotation with high accuracy across various species and tissues.


Single-cell RNA sequencing (scRNA-seq) is a technique that offers high-resolution analysis of transcriptional heterogeneity within a population of cells. Compared to bulk RNA sequencing (RNA-seq) that yields aggregated gene expression profiles, sc-RNAseq enables the identification of rare, and potentially novel, cell compartments within a diverse cell population. A typical scRNA-seq analysis pipeline involves dimensionality reduction techniques, which transform high-dimensional gene expression data into lower-dimensional latent spaces suitable for unsupervised clustering using algorithms like k-means clustering. Cell clustering coupled with community detection methods such as Louvain or Leiden, facilitate the identification of distinct cell populations (clusters) that can be represented in a two-dimensional latent space via uniform manifold approximation and projection. These clusters likely represent unique cellular states, highlighting both known cell types and potentially novel subpopulations that need further investigation. Deciphering the gene expression signatures of these cellular cluster’s allows researchers insights into cellular function, disease pathogenesis, and developmental trajectories.

A critical step in the scRNA-seq analysis is the identification of cell types, and subsequent annotation. Cell annotation has traditionally depended on manually identifying marker genes, a process that remains widespread due to the lack of single cell reference atlases for certain tissues or species, despite the availability of automated tools. The development of generative pre-trained transformers (GPTs) marks a significant shift towards greater efficiency and accuracy in bioinformatics.

A recent study published in Nature Methods by Hou W & Ji Z. (2024) highlights GPT-4’s ability to accelerate cell type annotation with high accuracy.

Quick and Accurate: GPT-4’s Efficiency in Cell Identification

The use of GPTs to perform cell annotation in sc-RNAseq analysis promises a future with minimal manual intervention, thereby streamlining the process considerably. Hou W. & Ji Z. (2024) wrote an R package called GPTCelltype, which facilitates an interface with GPT models, to evaluate GPT-4’s performance on ten cross-species, cross-tissue datasets, including both normal and cancer samples. They compared GPT4 to its predecessor, GPT-3.5, and other automated annotation tools such as CellMarker2, SingleR  and ScType . GPT-4 showed high accuracy, with its results matching or closely aligning with manual annotations in over 75% of cases. Moreover, GTP-4 distinguished between pure and mixed cell types with 93% accuracy and differentiated known from unknown cell types with 99% accuracy. Its robust performance across various species, tissues, and cell conditions highlights its versatility in the life sciences field.

Despite the high accuracy, there may be inconsistencies observed between manual and GPT4 based cellular annotation. For instance, cell types classified as stromal cells include fibroblasts and osteoblasts expressing type I collagen genes, and chondrocytes expressing type II collagen genes. Therefore, for cells manually labeled as stromal cells, GPT-4 would assign annotations with more granularity, such as fibroblasts and osteoblasts, and therefore result in partial matches and a lower agreement with manual annotations.  Although this agrees with the pattern observed in cells manually annotated as chondrocytes, fibroblasts, and osteoblasts, inconsistencies may arise with manual annotations with less granularity. In addition, GPT-4’s reliance on its training data and the necessity for human validation in certain cases highlight areas for caution. The study also points out the limitations related to high noise levels and the potential for AI hallucinations, which is a situation where an AI model generates incorrect or misleading information. These findings emphasize the importance of expert oversight when utilizing GPT-4 for single cell annotation. .

Next generation annotation

The use of GPT-4 in GPTCelltype for high accuracy cell annotation in sc-RNAseq datasets marks a significant step forward in the life sciences, particularly for research and development professionals in pharmaceutical companies. This technology has the potential to  speed up the early stages of drug discovery and development, particularly in the immunology space, accelerating cell type annotation which is often a bottleneck in sc-RNAseq analysis frameworks.  However, the integration of GPT-4 also necessitates careful consideration. As with any new technology, it is important to understand the biases, as well as the reasoning behind conclusions that can be derived from GPT-4. Developing robust validation processes and integrating human expertise alongside GPT-4 will be vital for harnessing its full potential while maintaining scientific rigor.

Outsourcing Bioinformatics Analysis: How Bridge Informatics (BI) Can Help

At Bridge Informatics, we understand the transformative impact of technologies like GPT-4 on the life sciences industry. BI’s data scientists prioritize studying, understanding, and reporting on the latest developments so we can advise our clients confidently. Our bioinformaticians are trained bench biologists, so they understand the biological questions driving your computational analysis. From pipeline development and software engineering to deploying your existing bioinformatic tools, BI can help you on every step of your research journey. As experts across data types from leading sequencing platforms, we can help you tackle the challenging computational tasks of storing, analyzing and interpreting genomic and transcriptomic data. Click here to schedule a free introductory call with a member of our team.

Shahrzad Ghazisaeidi, PhD, Data Scientist, Bridge Informatics

Shahrzad specializes in high-throughput sequencing, data pre-processing, and downstream analysis of transcriptomic and epigenetic landscapes. She is particularly passionate about developing innovative tools for drug repurposing.

Prior to joining Bridge Informatics, Shahrzad served as a Postdoctoral Associate at the Hospital for Sick Children in Toronto, Canada where she researched the epigenetics of peripheral nerve injury models.

Shahrzad earned her Ph.D. in Physiology and Neuroscience from the University of Toronto. Her studies focused on the sex-dependent and independent gene regulation of peripheral nerve injury. Currently based in Toronto, Shahrzad balances her professional pursuits with personal interests by making time for yoga, martial arts, and poetry.

Haider M. Hassan, Data Scientist, Bridge Informatics

Haider is one of our premier data scientists. He provides bioinformatic services to clients, including high throughput sequencing, data pre-processing, analysis, and custom pipeline development. Drawing on his rich experience with a variety of high-throughput sequencing technologies, Haider analyzes transcriptional (spatial and single-cell), epigenetic, and genetic landscapes.

Before joining Bridge Informatics, Haider was a Postdoctoral Associate at the London Regional Cancer Centre in Ontario, Canada. During his postdoc, he investigated the epigenetics of late-onset liver cancer using murine and human models. Haider holds a Ph.D. in biochemistry from Western University, where he studied the molecular mechanisms behind oncogenesis. Haider still lives in Ontario and enjoys spending his spare time visiting local parks. If you’re interested in reaching out, please email [email protected] or [email protected]

Share this article with a friend

Create an account to access this functionality.
Discover the advantages