Research 

Systems biology aims at understanding the design principles and multi-level properties of large cellular subsystems arising from numerous molecular interactions. While recent technological advancements has enabled the rapid collection of data on the molecular components of cells and their interactions, there is an increasing need for automated methods that can extract useful knowledge from these data and build computational models that accurately describe both normal cellular physiology and the phenotypic impact of mutations and environmental perturbations (e.g., drug treatments). The unicellular yeast, Saccharomyces cerevisiae, is an ideal candidate for systems biology studies due to the availability of large and diverse sets of post-genomic information and experimental tools. We are developing novel computational methods to analyse functional genomic datasets and to automate scientific discovery in the fields of systems biology and drug discovery by focusing on the following research topics:

i) Understanding genetic interaction networks

Why most single gene deletions do not show a lethal phenotype? How mutations in different genes interact to enhance or suppress the phenotype? How common are genetic interactions? What is the functional role of genes with an especially large number of genetic interactions? Answers to these questions have relevance not only to functional genomics, but also to problems such as which mutational paths are accessible for evolution and how deleterious mutations are eliminated from the population. With the recent availability of systematic genetic interaction maps in yeast, we are in a position to gain new insights into the above issues. In particular, we use data mining methods to integrate information on genetic interactions with other types of omics data (e.g., gene expression, protein-protein interactions, etc.) to infer cellular pathways and modules. We also investigate the mechanistic cause and evolution of gene dispensability (i.e., the apparent lack of growth defect after gene removal) by analysing single and double gene deletion phenotypes under different environmental conditions. For example, we demonstrated that a large fraction of gene pairs that can compensate mutations in each other under standard laboratory conditions display non-redundant functions under some other conditions. This finding supports the view that functional redundancy among genes is more apparent than real.

A conceptual model to explain conditional synthetic lethal genetic interactions in metabolism. A key metabolite (yellow circle) can be synthesized via three independent pathways. Metabolic genes A and B show synthetic lethality in Environment I, where starting nutrients of both pathways are present in the  medium. However, B is unable to compensate deletion of A in Environment II, and the double mutant is rescued by the third pathway in Environment III.

ii) Automated refinement of metabolic network models

Genome-scale metabolic models give a mechanistic mapping between genotype and phenotype and can be used to understand the behaviour of gene networks, to design novel strains for metabolic engineering applications, and to examine the process of genome evolution. The construction and refinement of such models, however, is largely performed manually, which is slow, difficult to reproduce, prone to biases, and cannot make efficient use of all types of high-throughput data. Automating model inference from systematic phenotype data is therefore of paramount importance. We are applying machine learning techniques to improve the metabolic network model of the yeast Saccharomyces cerevisiae, based on an unprecedented set of quantitative phenotypic data on millions of mutants (published single gene deletion data and unpublished double gene deletion data provided by our collaborator, Charles Boone, Toronto).

iii) Automated discovery of optimal antimicrobial drug combinations

Evolution of antimicrobial drug resistance is a problem that continues to challenge the healthcare industry. In addition to discover new compounds, there is an increasing need to identify optimal combinations of existing drugs (‘drug cocktails’) that are highly effective against resistant strains. Most high-throughput experimental screens aim to test all possible pair-wise combinations in a given drug compound library, which is not feasible for large libraries, or when the goal is to test the synergistic effects between three or more compounds. Therefore, robotic protocols coupled with intelligent experimental selection are needed to explore the vast chemical space of drug compounds in rapid and cost-effective ways. We are developing a heuristic algorithm that would optimize the composition of antimicrobial drug cocktails by iteratively performing experiments, using automated laboratory equipments, and automatically evaluating them. The project is run in collaboration with the Pál lab (BRC, Szeged).