Proteome predicts and interprets the metabolome of the cell by means of artificial intelligence

Speaker: Aleksej Zelezniak, Chalmers University of Technology, Gothenburg

Aleksej Zelezniak obtained Master’s degree in Biotechnology at Technical University of Denmark. He did his PhD EMBL-Heidelberg under supervision of Dr Kiran Patil. Aleksej is interested in development of novel ways of analysis and function and operation of metabolic networks. He combines multiple experimental data with modelling, statistical and machine learning approaches to study metabolism regulation. He recently finished his postdoc at Markus Ralser group at the University of Cambridge and the Francis Crick Institute in London. At the present he holds an assistant professor position at the Chalmers University of Technology, Gothenburg, Sweden where he is establishing his independent research group.

A key obstacle in solving the genotype-phenotype problem represent metabolite concentrations, which so far are not predictable even when genome, transcriptome or proteome of a cell are known. By systematically creating precise proteomes of yeast gene-knockouts strains, we noticed that the absence of each protein kinase triggers a highly specific reconfiguration in the global network of enzyme abundance. Interpreting these changes, we find that gene expression regulation re-distributes the control over metabolism between different sets of enzymes. Gene expression changes can hence influence metabolites without being directly correlated to them. To map the multifactorial interactions, we develop a strategy to associate enzyme and metabolite levels by the combinatorial application of machine learning. We succeed to dissect a genome spanning metabolism regulatory system in which every kinase plays a specific role, and achieve for the first time, the prediction of a complex eukaryotic metabolome out of enzyme expression information. Finally, by incorporating the metabolic network topology, our artificial intelligence approach is rendered interpretable and identifies key kinase-, enzyme-, and metabolite determinants of the yeast cell’s metabolic phenotype. Our study reveals that the hierarchical control of metabolism is regularly achieved through redistribution of flux control, revealing a fundamental principle how metabolic regulation is achieved. Further, we demonstrate that a new generation of data-driven biology can identify biological mechanisms underlying the genotype-phenotype problem. By rendering the complex phenotype – the cellular metabolome – quantitatively predictable, we picture the global role of enzyme abundance in metabolic regulation for achieving the phenotype of the cell.

Venue: Gamma 2, lunch room, SciLifeLab Solna