Advanced Machine Learning techniques for quantifying uncertainty in climate impact on biological systems

Climate change is affecting ecosystems globally (see for example https://acsess.onlinelibrary.wiley.com/toc/19403372/2024/17/1 and https://www.sciencedirect.com/science/article/pii/S1360138521003216), with rising temperatures, altered precipitation patterns, and extreme weather events inducing substantial stress on biological systems. Recent advances in high-throughput gene expression profiling technologies, such as RNA sequencing (RNA-seq) have revealed that these climate stressors can induce changes in gene expression patterns across a range of species. Such changes can occur as shifts in transcriptional programs, activation of stress response pathways, and modifications of metabolic processes, which may affect growth, reproduction, immune responses, and survival of different organisms.

Changes in gene expression are a critical aspect of how organisms respond to environmental stressors, serving as a molecular indicator of biological adaptation or stress. Persistent changes in gene expression can affect growth, development, reproductive success, and overall resilience of populations. This has direct implications for agriculture, biodiversity conservation, and ecosystem services.

The project’s focus on probabilistic modeling and network analysis provides a powerful means of mapping these complex interactions, contributing to a more comprehensive understanding of climate-biology dynamics. The project uses gene expression analysis, network estimation, and probabilistic approaches to model biological responses, to climate stressors. This work bridges biology, data science, and climate studies to address real-world challenges.

This PhD project focuses on quantifying uncertainties in climate impacts on biological systems such as plant stress responses, animal reproductive health, and marine ecosystem adaptations using advanced statistical and machine learning techniques. The research leverages gene expression data from publicly available repositories like the Gene Expression Omnibus (GEO). Initial datasets include GSE252418 (heat-driven epigenomic changes in mice), GSE206753 (grape response to heatwaves), and GSE152444 (marine life adaptation markers), capturing diverse biological responses to environmental stress.

Research Focus:

The primary goal is to develop probabilistic models, including Gaussian graphical models and Bayesian frameworks, to understand the complex relationships between climate variables and biological responses. By characterizing variability and uncertainty, these models will map gene and protein network changes under climate stress, identifying pathways of resilience or vulnerability.

Impact and Methods:

Students will gain hands-on experience with statistical modeling, network analysis, and machine learning, enhancing skills critical for data science, bioinformatics, and climate biology. Decision-support tools created during the project will quantify uncertainties in predicted responses, aiding policymakers in risk assessments and adaptive climate strategies. Collaborative opportunities include integration with the UNRISK CDT’s extensive network, fostering interdisciplinary approaches to quantifying climate risks and uncertainties.

This PhD is well-suited for students with a strong foundation in mathematics, statistics, or computer science who are eager to apply their skills to pressing issues in climate and biological sciences. Ideal candidates will have experience in statistical modeling, machine learning, or data science, with an interest in environmental or biological applications. Applicants should be motivated by complex, interdisciplinary challenges and have a desire to work with high-dimensional datasets to understand how climate variability impacts biological systems. Familiarity with probabilistic methods and/or network analysis is advantageous but not required, as comprehensive training will be provided.