CancerTrace: Multi-stage single-cell analysis of networked cancer evolution for driver and modulator gene...

Area of Science:

• Computational Biology
• Genomics
• Cancer Research

Background:

• Identifying patient-specific cancer drivers is difficult due to temporal changes and limited data.
• Current methods often use static correlations or broad modules, missing dynamic, causal driver-modulator links.

Purpose of the Study:

• To introduce CancerTrace, a novel computational framework for uncovering time-resolved, patient-specific cancer driver genes and their regulatory mechanisms.
• To overcome limitations of existing methods by analyzing single-cell RNA sequencing data dynamically.

Main Methods:

• CancerTrace integrates Transfer Entropy and sparse conditional structure within a variational Bayesian model.
• It analyzes longitudinal single-cell RNA sequencing (scRNA-seq) data to reconstruct stage-resolved expression dynamics.
• The framework maps directed influences from modulators to drivers without needing DNA or perturbation assays.

Main Results:

• CancerTrace identified an EpCAM-anchored epithelial compartment in lung adenocarcinoma (LUAD) patient data.
• It recovered known and novel drivers, including VPS37D and ATP11AUN, and validated known driver-gene interactions.
• The framework demonstrated measurable influence of epithelial drivers on NK cells, reflecting biological changes over time.

Conclusions:

• CancerTrace effectively infers causal, time-directed driver-modulator relationships from scRNA-seq data.
• This approach overcomes static and cohort-dependent limitations, offering a powerful tool for mechanistic understanding in precision oncology.