Deconformer: efficient cell-type composition profiling of cfRNA across diverse cell types.

Deconformer accurately estimates cfRNA contributions across 60 human cell types with lower computational cost than existing methods.

• Why it matters: Understanding cell-type contributions in cfRNA is crucial for liquid-biopsy research, but current methods face challenges in profiling many cellular sources efficiently.
• What they did: Deconformer employs a biologically informed deep-learning framework with a Transformer architecture and pathway knowledge, tested on simulated and real cfRNA data.
• The result: It demonstrated improved robustness and accuracy, recapitulated known biological trajectories, and detected disease-related shifts, enabling scalable cfRNA profiling.