Developing Conditional Independence Tests for Causal Discovery with Multiple (Co)Variance Components and Heterogeneous Variables: Application to Understanding Variability in Malaria Risk in the Brazilian Amazon

• Grant number: 25/06122-8
• Support Opportunities: Research Grants - Visiting Researcher Grant - International
• Start date: July 27, 2025
• End date: August 10, 2025
• Field of knowledge: Physical Sciences and Mathematics - Probability and Statistics - Applied Probability and Statistics
• Principal Investigator: Júlia Maria Pavan Soler
• Grantee: Júlia Maria Pavan Soler
• Visiting researcher: Adele Helena Ribeiro
• Visiting researcher institution: Westfälische Wilhelms-Universität Münster/Wwu, Germany
• Host Institution: Instituto de Matemática e Estatística (IME). Universidade de São Paulo (USP). São Paulo , SP, Brazil
• Associated research grant: 22/11963-3 - Individual variation in malaria risk: causes and consequences in Amazonian populations, AP.TEM

Abstract

The primary goal of the visit is to advance the development of a methodological framework for causal discovery from observational data exhibiting complex dependency structures and diverse variable types. Building on our previous collaboration - focused on causal inference using conditional independence tests based on linear mixed models for analyzing Gaussian data from family-based studies - our project aims to extend these methods to accommodate the multimodal, heterogeneous, and non-i.i.d. nature of data encountered in longitudinal genetic-epidemiological research. During the visit, we will refine the proposed statistical methodology, implement novel conditional independence tests, and evaluate their performance using simulated datasets that capture the key challenges of our real-world application: the analysis of malaria risk variability in the Brazilian Amazon. These challenges include temporal, spatial, genetic, and household-level dependencies; a variety of variable types (e.g., continuous, binary, multinomial, count); and complex patterns of missing data. This collaboration will leverage our complementary expertise in causal discovery, statistical modeling, and human genetics to lay the foundation for a robust analysis of the Mâncio Lima dataset. The visit will also include participation in the RBras meeting, held from August 4 to 8, 2025, in Vitória ES, where we will jointly present a short course entitled "Aprendizagem e Inferência Causal: Um Guia Prático" and a poster titled "Simulações sobre o Impacto de Parâmetros Genéticos e Ambientais na Variabilidade do Risco de Malária" (AU)