Pertpy Links Analysis and Interpretation
pertpy was developed by a team at the Institute of Computational Biology at Helmholtz Munich and at the TUM, led by Prof. Fabian Theis. First author Dr. Lukas Heumos, a scientist in the Theis lab, designed the framework as a building block in an international open-source platform for single-cell analysis. To allow researchers to analyse single-cell perturbation experiments in a seamless workflow, pertpy reads data from these experiments in a standard single-cell format, enriches them with knowledge from public databases and provides statistical methods in a unified interface. In this way, even very large datasets from single-cell perturbation experiments can be analysed with a few, reproducible workflows – up to questions such as which cell types and gene programs respond most strongly to a given intervention.
Finding Patterns in Complex Datasets
In perturbation experiments, cells often show complex response patterns: some cell types react strongly, others hardly at all; entire groups of genes are up- or downregulated at the same time. The goal for researchers is to identify and interpret these patterns: Which cell types are particularly sensitive? Which gene programs respond in concert, and how similar are the effects of different perturbations?
With pertpy, researchers can now answer such questions within a single workflow and turn an initially confusing cloud of data into a clearer picture. “We have reassembled widely used methods in a modern, easy-to-use form,” says Lukas Heumos. Researchers can combine the individual modules like building blocks into custom analysis pipelines instead of fighting their way through large numbers of incompatible tools. “In many cases we didn’t need to develop new algorithms at all – we simply had to connect existing tools in a clean way,” Heumos adds.
Road-Tested: Pertpy Extracts More Insight From Data
A particular strength of pertpy is its tight integration with publicly available databases. For example, the software can automatically retrieve information on cell lines, tumour entities, compounds and their targets and link them to the measured single-cell data. A simple numeric matrix thus becomes a biologically annotated dataset: researchers not only see that the activity of a gene changes, but also get hints as to which pathways, cell lines or mechanisms of action these changes may be related to.
The team has demonstrated how pertpy can be used in practice on several different datasets. In one large-scale experiment with leukaemia cells, researchers used the CRISPR gene-editing system to switch individual genes on or off. With pertpy, the researchers was able to reveal additional gene programs involved that had remained hidden in the original analysis – for example pathways that have similar effects on cell proliferation but act through distinct molecular mechanisms.
Another example comes from the clinic. Here, immune cells from tumour biopsies of patients with an aggressive form of breast cancer, triple-negative breast cancer, were analysed before and after treatment as part of chemo- and immunotherapy. In this dataset, pertpy could show how specific T cell subsets and shared gene programs change over the course of treatment – and which patterns are associated with response to therapy.
“Perturbation data are among the most informative, but also the most complex datasets in biomedicine,” says Fabian Theis. “With pertpy, we can translate this complexity into clearer questions: Which cell types respond most strongly? Which gene programs are co-regulated? And what does this mean for the development of new therapies?”
An Open Community Project
In the long term, pertpy is set to grow further: beyond standard single-cell RNA-seq data, the team plans to incorporate image-based readouts from large-scale cell screens as well as combined multi-omics experiments, in which, for example, gene expression and chromatin accessibility are measured in the same cells. “We see pertpy as an open community project,” says Theis. “The more labs contribute their methods and data types, the better we can exploit perturbation data – up to AI models that predict how cells will respond to specific interventions.”
About Pertpy
pertpy is freely available, open source and is being further developed by an international community – new methods and data types can be integrated step by step. To make it easier for users to get started, the team provides extensive online documentation alongside the software itself, as well as numerous step-by-step tutorials: https://pertpy.readthedocs.io/en/latest/index.html
Original publication
Heumos et al., 2025: Pertpy: an end-to-end framework for perturbation analysis. Nature Methods. DOI: 10.1038/s41592-025-02909-7
