Pierpaolo Fiore

Single-cell RNA sequencing enables high-resolution characterization of transcriptional heterogeneity, but provides only a partial view of the regulatory and metabolic processes associated with cellular states. Several computational methods infer transcription factor (TF) activity and metabolic features directly from RNA, yielding complementary functional representations of cellular organisation. Here, we use a latent factor organisational strategy to jointly model four transcriptome-derived functional projections—gene expression, TF regulon activity, metabolite-level features and predicted metabolic fluxes. Although all layers originate from the same measurement, each captures distinct regulatory or metabolic programs. The resulting latent space organizes these inferred programs into coordinated axes of variation guided by complementary regulatory and metabolic constraints, facilitating functional interpretation beyond gene expression alone. When applied to breast cancer cell line data, the proposed framework organises transcriptome-derived signals into distinct functional programs, including proliferative, oxidative-metabolic and stress-associated axes, that are only partially resolved in RNA-only analyses. Overall, we demonstrate that regulatory and metabolic programs inferred from scRNA-seq can be structured into an interpretable latent representation, supporting a more coherent functional characterization of cellular states from transcriptome-derived functional projections.

Unbiased learning pipeline for label-free single-cell classification.