Description
Instructors: Kynon J Benjamin, Minoli Perera
Analytic decisions in human genetics – such as ancestry representation, covariate selection, and cross-population result aggregation – can shape biological discovery and health disparities but are often treated as purely technical. This interactive workshop demonstrates how routine modeling choices affect inference in genome-wide and multi-omic studies of diverse populations through real-time exploration of simulated data. Emphasizing interpretation over software, it equips participants with a practical framework for parity-aware omics analysis and critical evaluation of analytic choices.
Session Details:
- Overview of analytic gaps and motivation: This opening segment introduces key gaps in current genomic analysis practices related to ancestry-aware inference. [15 minutes]
- Methodology and platform orientation: Participants are introduced to the web-based interface and the structure of the simulated datasets. [15 minutes]
- Interactive exercises I: ancestry representation and association testing: Participants explore scenarios involving global versus local ancestry adjustment, pooled versus stratified analyses, and within-ancestry versus meta-analytic approaches. Real-time visualizations demonstrate how these choices alter effect estimates, power, and interpretability. [30 minutes]
- Interactive exercises II: regularization, predictability, and scalability: Participants examine scenarios where reduced model performance reflects meaningful biological heterogeneity rather than noise, and explore how scalable GPU-accelerated frameworks enable genome-wide inference. [30 minutes]
- Full-group synthesis and discussion: The group reconvenes to synthesize insights from the interactive exercises. [15 minutes]
- Key takeaways, best-practice framework, and Q&A. [15 minutes]
Learning Objectives:
- Identify analytic decisions (e.g., ancestry, covariates, regularization, meta-analysis) that affect inference in diverse populations.
- Evaluate how alternative analytic frameworks influence statistical power, interpretability, and discovery in multi-ancestry studies.
- Interpret changes in model performance and predictability as signals of biological heterogeneity, not just analytic success or failure.
- Apply a parity-aware analytic framework to assess tradeoffs and assumptions in genetic and genomic study design
Additional Information:
Basic level; Delivered through a browser-based platform requiring no local installation or coding.