BifurcatoR: A Framework for Revealing Clinically Actionable Signal in Variance Masquerading as Noise

Area of Science:

• Biomedical data analysis
• Computational biology
• Statistical genetics

Background:

• Disease heterogeneity complicates medical research and treatment.
• Standard analyses often assume population homogeneity, missing biologically distinct patient subgroups.
• Variance heterogeneity (VH) offers a lens to detect latent etiological structures for prognosis and therapeutic response.

Purpose of the Study:

• To develop an accessible software platform for detecting, modeling, and interpreting VH.
• To address limitations of existing VH methods, including reliance on normality assumptions and need for programming expertise.
• To provide guidance on study design for VH analysis.

Main Methods:

• Developed BifurcatoR, an open-source software platform with a web interface.
• Integrated simulation-based method evaluation and study design recommendations.
• Benchmarked VH methods via simulation and applied BifurcatoR to acute myeloid leukemia (AML) and obesity datasets.

Main Results:

• VH method performance is context-specific, depending on data distribution and subgroup structure.
• Identified two distinct AML subgroups with differential treatment responses, including a poorer prognosis EVI1-high group.
• Uncovered immunophenotypic subgroups in obesity linked to adipose immune cell composition variation.

Conclusions:

• VH represents structured, clinically relevant biological signals, not noise.
• BifurcatoR provides a practical framework for integrating VH into biomedical research.
• VH analysis has implications for biomarker discovery, patient stratification, and precision medicine.