Automating Hypothesis Generation and Testing: Towards Self-driving Biology

Doktorsavhandling, 2025

Biological systems remain only partially understood, and the relative pace of functional discovery has been slowing down despite advances in measurement technologies. A growing consensus suggests that the most promising way forward is not only via conventional laboratory automation, but through the development of fully autonomous systems that can generate, prioritize, draw insight from, and execute high-throughput experimentation.

This thesis explores how such automation can accelerate scientific discovery by combining methods from artificial intelligence—such as inductive logic programming, explainable AI, and large language models—with physical instrumentation, including laboratory robotics and high-throughput analytical platforms like mass spectrometry. The work spans the entire discovery cycle, from hypothesis generation to experimental evaluation.

As a case study, the methods are applied to Saccharomyces cerevisiae (baker’s yeast), an extensively studied eukaryote and a powerful model organism for systems biology. In doing so, the thesis contributes to further characterization of key aspects of yeast biology, including the diauxic shift and its regulators (via untargeted metabolomics), genome-wide proteomic regulation, phenotypic determinants of fitness, and metabolic interactions involving amino acids.

The findings emphasize that automation in biology requires more than throughput alone. Automated systems must also leverage existing knowledge, provide interpretable reasoning processes, and preferably capture enough metadata for auditability. These studies also highlight how automation, when combined with structured knowledge and high-throughput experimentation, can refine existing approaches and move biology toward more integrative and transparent modes of discovery.