Abstract
Recent advances in single-cell technologies enable scientists to measure molecular data at high-resolutions and hold the promise to substantially improve clinical outcomes through personalised medicine. However, due to a lack of tools specifically designed to represent each sample (e.g. patient) from the collection of cells sequenced, disease outcome prediction on the sample level remains a challenging task. Here, we present scFeatures, a tool that creates interpretable molecular representation of single-cell and spatial data using 17 types of features motivated by current literature. The feature types span across six distinct categories including cell type proportions, cell type specific gene expressions, cell type specific pathway scores, cell type specific cell–cell interaction scores, overall aggregated gene expressions and spatial metrics. By generating molecular representation using scFeatures for single-cell RNA-seq, spatial proteomic and spatial transcriptomic data, we demonstrate that different types of features are important for predicting different disease outcomes in different datasets and the downstream analysis of features uncover novel biological discoveries.
Competing Interest Statement
The authors have declared no competing interest.