"Integrative computational approaches for modeling complex disease biology across scales and systems"

Abstract

To effectively model the molecular underpinnings of complex traits and diseases, computational methods must integrate diverse data types, handle partial or limited observations, and remain robust to variations in dataset size. In this talk, I will present several recent methods developed to address these challenges across diverse studies, assay types, and organisms, leveraging novel statistical and machine learning approaches. 

First, I will introduce ALPINE, an NMF-based framework that disentangles the influence of technical and non-relevant phenotypic factors in single-cell transcriptomic data, enabling the integration of multiple studies. 

Integrating across data types, I will discuss our method, seismic, which combines genome-wide association studies with single-cell RNA sequencing to prioritize disease-relevant cell types, linking genetic variation to cellular function. 

Finally, I will discuss ETNA, a machine translation-inspired approach that embeds protein-protein interaction networks from different organisms into a shared space, facilitating cross-species functional comparisons. 

Together, these methods highlight how diverse data sources can be integrated across molecular, cellular, and organism levels to better model complex disease biology.