CytoVI unifies analysis of antibody-based single-cell data

Nature Methods has reported CytoVI, a computational model designed to bring antibody-based single-cell datasets into a common analytical framework. The software is intended to address problems that can make results difficult to compare, including technical noise, batch effects, differences between platforms and the limited number of antibodies available in individual panels.

CytoVI combines several analytical tasks in one probabilistic model. It produces cell embeddings that capture informative patterns in the data, estimates measurements that were not observed, tests for differences in protein expression and supports automated cell annotation. The model was developed for data generated by technologies including flow cytometry, mass cytometry and CITE-seq, all of which are used in clinical diagnostics and biological research.

A common framework for different measurements

Antibody-based single-cell methods do not necessarily examine the same proteins or produce results in the same format. A restricted antibody panel can leave important measurements unavailable, while technical variation and batch effects can obscure biological signals. Differences between platforms add another complication when researchers try to analyse datasets together.

According to Nature Methods, CytoVI uses a single statistical framework to handle these sources of complexity rather than treating each analytical step as a separate process. Its cell embeddings provide a way to represent cells using the patterns detected across the measurements. The model can also impute missing values, allowing absent measurements to be estimated within the analysis.

The same framework is used for differential protein expression testing and cell annotation. This is intended to make it possible to examine protein-level differences and assign cells to types or states without relying on a sequence of disconnected tools. The approach is described as probabilistic, meaning that the model represents uncertainty as part of its analysis rather than presenting every result as entirely fixed.

Testing CytoVI on B-cell development

The researchers applied the model to create an integrated atlas of B-cell maturation. The atlas brings together measurements covering 350 proteins, giving CytoVI a broad dataset with which to model changes across B-cell development.

That analysis identified proteins linked with immunoglobulin class-switching. Class-switching is represented in the study as a biological process associated with particular protein patterns, and the model was used to identify the proteins connected with it. The result demonstrates how the system can be used not only to combine datasets but also to find protein associations within a larger cellular map.

The atlas is also an example of the type of analysis that becomes more difficult when data come from different experiments or platforms. By placing measurements in an integrated model, CytoVI was used to examine maturation across a wider protein range than an individual restricted panel might provide.

Findings in lymphoma samples

CytoVI was also assessed using samples from patients with B-cell non-Hodgkin lymphoma. These samples were profiled with both flow cytometry and CITE-seq, allowing the model to be used across two antibody-based single-cell technologies.

In that patient cohort, the analysis revealed T-cell states associated with the disease. The finding shows that the method can identify changes beyond the B-cell population named in the cancer diagnosis. It also illustrates the use of cross-platform analysis in a clinical disease setting, where measurements made with different technologies need to be considered together.

The reported applications position CytoVI as a tool for both atlas construction and disease-focused investigation. Its functions cover representation of cells, recovery of missing measurements, protein-expression comparisons and annotation, while its probabilistic design is intended to account for uncertainty and technical variation within one model.

CytoVI has been released as open-source software through scvi-tools.org. Its availability gives researchers a way to apply the method to antibody-based single-cell data in their own clinical or research analyses. Nature Methods presents the software as a unified approach for working with datasets affected by platform differences, batch effects, noise and incomplete antibody panels.