Main project

K-CLIER

Interpretable dimensionality reduction for kidney biology

• Life sciences
• Healthcare

K-CLIER is trained on a single-cell atlas and a large bulk RNA-seq data set, then applied to independent data sets in renal disease. Graphical abstract from Legouis et al., iScience 2024 (open access).

Challenge

Single-cell transcriptomics has refined our understanding of the cellular composition of tissues and organs. Its use in clinical practice, however, faces major challenges related to tissue processing and data analysis, while bulk transcriptomics is easier to obtain but harder to interpret at the level of cell types and states.

Approach

K-CLIER builds on PLIER, a matrix factorisation method that aligns latent variables with prior gene sets. Using transfer learning, it learns a transformation of bulk gene expression into a low-dimensional representation whose latent variables correspond to cell types and states of a single-cell atlas. Prior biological knowledge, in the form of single-cell signatures, therefore guides the construction of the representation.

Results

We applied the transformation to data from patients with renal diseases. The resulting variables were used to predict clinical outcomes and therapy response in oncological patients and to study rejection in kidney transplantation, and provided evidence for a central role of altered proximal tubule cells in kidney fibrosis. The approach extends to other clinical contexts. A federated version of the underlying method (FPLIER) allows pathway-level analysis of data distributed across institutions.

Publications

• A transfer learning framework to elucidate the clinical relevance of altered proximal tubule cell states in kidney disease

  D. Legouis, A. Rinaldi, D. Malpetti, G. Arnoux, T. Verissimo, A. Faivre, F. Mangili, A. Rinaldi, L. Ruinelli, J. Pugin, S. Moll, L. Clivio, M. Bolis, S. de Seigneux, L. Azzimonti, P.E. Cippà

  iScience, 27(3), 109271, 2024

• Protocol for interpretable and context-specific single-cell-informed deconvolution of bulk RNA-seq data

  D. Malpetti, F. Mangili, M. Bolis, A. Rinaldi, D. Legouis, L. Ruinelli, P. Cippà, L. Azzimonti

  STAR Protocols, 6(1), 103670, 2025Code

• FPLIER: Federated Pathway-Level Information Extractor

  D. Malpetti, C. Berchtold, F. Gualdi, M. Scutari, L. Azzimonti, F. Mangili

  ACM Conference on Bioinformatics, Computational Biology and Health Informatics (ACM BCB 2026), 2026Code

Software

• K-CLIER protocol

  Step-by-step protocol, with R code and test data, for interpretable, single-cell-informed deconvolution of bulk RNA-seq data.

  GitHubSTAR Protocols

• FPLIER

  Federated version of PLIER, which extracts pathway-level representations of transcriptomic data distributed across institutions, with a simulation based on the Flower framework.

  GitHubPaper