Learning the language of the genome: AI- and data-driven decoding of genetic variants and disease mechanisms

Genetic variation plays a crucial role in human disease susceptibility and molecular pathophysiology. This thesis explores genetic determinants of disease using advanced computational approaches, including statistical modeling, network analysis, and deep learning.

Chapter one introduces key concepts in genetic variation, disease mechanisms, and computational methodologies.

In chapter two, we investigate inborn errors of type I interferon immunity as a factor in severe acute hepatitis E, identifying damaging variants in viral sensing and interferon signaling genes that contribute to symptomatic infection.

Chapter three presents machine learning frameworks for rare disease gene discovery. Using a genomic language model, we generate gene embeddings that identify causal genes and pathways, validated in rare disease patients.

In chapter four, we fine-tune the ESM2 protein language model to predict the functional impact of missense variants, facilitating variant interpretation and classification.

In chapter five, we develop proteome-wide predictive models to classify disease-associated proteins by mode of inheritance and molecular mechanism, distinguishing autosomal dominant from recessive proteins and categorizing dominant proteins into loss-of-function, gain-of-function, or dominant-negative.

Chapter six introduces NMDEP, a model for predicting nonsense-mediated mRNA decay (NMD) efficiency. By integrating optimized rule-based heuristics, sequence embeddings, and biological features, NMDEP achieves state-of-the-art transcript degradation predictions and identifies key NMD determinants.

Finally, chapter seven discusses the potential and limitation of AI- and data-driven approaches in decoding genetic disease mechanisms, with implications for precision medicine, therapeutic target discovery, and disease prevention.