studyStrap

studyStrap implements ensemble methods that combine hierarchical resampling ("study straps") and covariate-profile similarity weighting to improve predictive generalizability across heterogeneous multi-study datasets.

Key Features:

• Covariate-Profile Similarity Weighting: Incorporates covariate similarity between training studies and target validation data to weight ensemble members.
• Hierarchical Resampling Scheme — Study Straps: Generates pseudo-study replicates by resampling across multiple studies using a generalized randomized cluster bootstrap to preserve multi-study structure.
• Tuning Parameter Control: Uses a tuning parameter to set the proportion of observations drawn from each study during resampling, interpolating between single-study sampling and pooled standard bootstrap.
• Ensemble Weighting Scheme: Applies ensemble weights informed by the distribution of covariates in the test dataset to enhance generalization to the validation study.

Scientific Applications:

• Real-time neurochemical sensing and awake neurosurgery: Estimating neurotransmitter concentrations from electrical measurements recorded during awake neurosurgery and related real-time neurochemical sensing experiments in humans.
• Multi-study predictive modeling for replicability: Improving replicability and generalizability of prediction models trained on heterogeneous datasets collected under varying conditions.

Methodology:

Generates pseudo-study replicates via a hierarchical resampling scheme ("study straps") using a generalized randomized cluster bootstrap; controls the study-wise sampling proportion with a tuning parameter; computes covariate-profile similarity between training studies and target validation data and incorporates that similarity into ensemble weighting using the test dataset covariate distribution.