From complexity to clarity: Researchers launch new tool to navigate method selection for single call datasets

Centre Research Leaders Associate Professor Pengyi Yang and Dr Shila Ghazanfar from the School of Mathematics and Statistics, together with colleagues at the Computational Systems Biology Unit at the Children’s Medical Research Institute, have published a new paper in Nature Methods. The new paper, titled Multi-task benchmarking of single-cell multimodal omics integration methods, introduces a decision-tree tool, and an interactive online resource to make method selection for single-cell data sets simple and effective.

New technologies have enabled scientists to measure multiple layers of information from single cells, including genes, proteins, and the packaging of DNA. These advances provide powerful insights into how cells work, but they also create complex datasets that are challenging to analyse. Many computational methods have been developed to combine these different types of data; however, it has been unclear which methods perform best for various research goals.

The paper, published in Nature Methods [date to be determined], is the first to systematically compare 40 of these computational methods across seven common analysis tasks, using 86 datasets. The study revealed clear trade-offs between preserving biological signals and removing technical noise, with deep learning methods often excelling but demanding more expertise and computing power. To guide researchers, the authors provide practical recommendations, a decision-tree tool, and an interactive online resource to simplify and enhance the effectiveness of method selection.

With potential applications in understanding cell identity, disease mechanisms, and advancing regenerative medicine, single-cell multimodal omics technologies are rapidly transforming biology. These technologies capture multiple layers of information from the same cell, from gene expression to protein abundance and epigenetics, offering an unprecedented view into how cells function.

“With this systematic categorisation and comprehensive benchmarking of current methods, this study sets the standard on future method development and also choosing the most appropriate method for single-cell multimodal omics data analysis’’ – Pengyi Yang

The findings highlight essential trade-offs among methods: some tools are stronger at preserving subtle biological signals, while others excel at filtering out technical noise. Deep learning methods often perform more competitively, but require greater expertise and computing power. By providing a decision-tree guide and an online resource, this work equips researchers worldwide to make informed choices, accelerating discoveries in cancer biology, regenerative medicine, and beyond.