SenPred: a single-cell RNA sequencing-based machine learning pipeline to classify deeply senescent dermal fibroblast cells for the detection of an in vivo senescent cell burden

• Published in: Genome medicine Vol. 17; no. 1; pp. 2 - 16
• Main Authors: Hughes, Bethany K., Davis, Andrew, Milligan, Deborah, Wallis, Ryan, Mossa, Federica, Philpott, Michael P., Wainwright, Linda J., Gunn, David A., Bishop, Cleo L.
• Format: Journal Article
• Language: English
• Published: London BioMed Central 14.01.2025; BioMed Central Ltd; Springer Nature B.V; BMC
• Subjects: 3D organotypic culture; Age; Analysis; Antibiotics; Bioinformatics; Biomedical and Life Sciences; Biomedicine; Cancer Research; Cell cycle; Cellular Senescence - genetics; Classification; Datasets; Fibroblasts; Fibroblasts - cytology; Fibroblasts - metabolism; Genomics; Human Genetics; Humans; Humidity; Immunofluorescence; In vivo senescence detection; Learning algorithms; Living skin equivalent; Machine Learning; Medicine/Public Health; Metabolomics; Morphology; Pipe lines; Pipelines; RNA; RNA sequencing; RNA-Seq; scRNA-seq; Senescence; Sequence Analysis, RNA - methods; Single-Cell Analysis - methods; Skin; Skin - cytology; Systems Biology; Transcriptome; Transcriptomics; In vivo senescence detection; Living skin equivalent; 3D organotypic culture; scRNA-seq; Senescence; Machine learning
• ISSN: 1756-994X; 1756-994X
• DOI: 10.1186/s13073-024-01418-0

Background Senescence classification is an acknowledged challenge within the field, as markers are cell-type and context dependent. Currently, multiple morphological and immunofluorescence markers are required. However, emerging scRNA-seq datasets have enabled an increased understanding of senescent cell heterogeneity. Methods Here we present SenPred, a machine-learning pipeline which identifies fibroblast senescence based on single-cell transcriptomics from fibroblasts grown in 2D and 3D. Results Using scRNA-seq of both 2D and 3D deeply senescent fibroblasts, the model predicts intra-experimental fibroblast senescence to a high degree of accuracy (> 99% true positives). Applying SenPred to in vivo whole skin scRNA-seq datasets reveals that cells grown in 2D cannot accurately detect fibroblast senescence in vivo. Importantly, utilising scRNA-seq from 3D deeply senescent fibroblasts refines our ML model leading to improved detection of senescent cells in vivo . This is context specific, with the SenPred pipeline proving effective when detecting senescent human dermal fibroblasts in vivo, but not the senescence of lung fibroblasts or whole skin. Conclusions We position this as a proof-of-concept study based on currently available scRNA-seq datasets, with the intention to build a holistic model to detect multiple senescent triggers using future emerging datasets. The development of SenPred has allowed for the detection of an in vivo senescent fibroblast burden in human skin, which could have broader implications for the treatment of age-related morbidities. All code for the SenPred pipeline is available at the following URL: https://github.com/bethk-h/SenPred_HDF .