AMS Special Seminar Series | Xiran Liu

Location: Shaffer 3

When: January 12th at 1:30 p.m.

Title: From variation to insight: revealing latent structure in high-dimensional biological data 

Abstract: High-dimensional biological datasets encode latent structure that links variation to mechanism. I will present three projects, motivated by distinct genetic applications, that examine latent structure to turn data variation into insight. 

Part I focuses on revealing meaningful structure by streamlining unsupervised discovery, comparison, and interpretation. Clumppling (Liu et al. 2024) aligns repeated population-structure inferences (mixed-membership clustering of individual genotypes) across clustering runs and parameters, using optimization and graph-mining to efficiently summarize alternative substructures. ACE-OF-Clust (submitted) extends these alignment ideas to single-cell and spatial transcriptomics, formalizing model comparison, introducing new metrics to identify clustering-informative features, and connecting shared structure across data types. 

Part II addresses how to characterize structure in complex correlated signals. ML-MAGES (Liu et al. 2025) investigates the genetic architecture of complex traits by controlling for correlation among variants with a neural network and using an infinite mixture model to extract multi-trait patterns that disentangle and categorize association signals. 

Together, these methods improve robustness, interpretability, and transferability across modern genomics workflows, and target data challenges that arise far beyond biology.

Bio: Xiran is a Postdoctoral Researcher at the Brown University Data Science Institute, working at the intersection of data science, population genetics, and computational biology. She develops innovative computational methods to interpret genetic variation, addressing critical challenges such as population admixture and the genetic architecture of complex traits. Her recent research explores how structured patterns reveal fine-scale relationships between populations, individuals, and genes. Xiran earned her PhD in Computational and Mathematical Engineering from Stanford University in 2023.

Zoom link: https://wse.zoom.us/j/92215066845