mlrMBO

mlrMBO optimizes subgroup-specific survival prediction by using model-based optimization to assign weights across cohorts and tune Cox proportional hazards models for high-dimensional genetic covariates.

Key Features:

• Subgroup-Based Weighted Likelihood: Assigns individual weights to observations from different patient subgroups to build prediction models for a target subgroup while accounting for cohort heterogeneity.
• Model-Based Optimization (MBO): Uses MBO to navigate hyperparameter space and to optimize per-observation subgroup weights for improved model performance.
• Cox Proportional Hazards Integration: Optimizes weights and hyperparameters within a Cox proportional hazards framework to produce subgroup-specific survival models.
• High-Dimensional Genetic Data Handling: Supports modeling with high-dimensional genetic covariates common in genomic and cancer research.
• Cohort Similarity Evaluation: Reflects similarity between cohorts by adjusting weights on a continuous scale (0–1) to balance variance reduction and bias control.

Scientific Applications:

• Oncology subgroup survival prediction: Enables development of survival models tailored to specific cancer subgroups using data pooled from multiple cohorts.
• Cross-cohort integration for small-sample studies: Improves predictive reliability when individual cohorts have limited sample sizes or high censoring by borrowing information across cohorts.
• Genomic and precision medicine analyses: Applies to studies with high-dimensional genetic covariates for precision oncology and biomarker-driven survival modeling.

Methodology:

Uses model-based optimization to determine optimal per-observation subgroup weights (scaled 0–1) and tune hyperparameters within a Cox proportional hazards framework to enhance subgroup-specific survival prediction across multiple cohorts.