Improving predictions in metabolic engineering problems by incorporating enzyme structural information

Systems biology foundations in this post-genomic era broadly rely upon well performed gene functional annotations. However, the experimental determination of a protein function is a laborious and expensive process, being its application unfeasible for the growing amount of sequences annotated. Current methodologies are based on computational tools that predict protein function over sequence similarity, but neglect a significant part of the putative proteins, ignore structural features and propagate annotation errors. The present thesis focuses in the improvement of predictions in metabolic engineering problems by incorporating enzyme structural information. The uncertainty on the usage of NADP(H) or NAD(H) as co-factors was addressed due to the major impact in metabolic engineering applications these molecules have, severely affecting both predictions and strain design results. The molecular determinants for cofactor specificity were unveiled, using enzyme structural information from a representative dataset of enzymes present in Protein DataBank with NAD(P)(H) as cofactors, and support vector machines. The integration of homology modelling tools and a support vector machine predictive model allowed a process automation for predicting cofactor specificity. The resulting software was made available online in the form of a webserver. Cofactor prediction of a curated dataset of structurally uncharacterized enzymes was performed with success, validating the developed method. The analysis and curation of the use of these cofactors in genome-scale metabolic models (GEM) was also performed through the cofactor prediction of sequences from 59 GEMs associated to reactions using NAD(P)(H). Results show some inconsistencies in GEM curation that can impair the correct simulation of the models, with an overall estimate of 28% of the genes implemented in the model being misclassified for cofactor specificity. The most recent GEM from Saccharomyces cerevisiae, Yeast 7.6, was corrected following the predictions performed and generally showed considerably improved results compared with the original version in the central metabolism, particularly for the flux in the pentose phosphate pathway. With the information on NAD(P)(H) cofactor specificity generated, a method was also developed and implemented to automatically predict the set of optimal gene mutations necessary for changing the NAD(P)(H) cofactor specificity of enzymes with unknown structure. Preliminary data shows promising results for the development of a tool for the automatic prediction of cofactor changing mutations, but experimental validation and further development are still a requirement.

Name thesis:degree_name

Doutoramento em Bioengenharia

Grantor

Universidade do Minho

Year dc:date.issued

2018

Author dc:creator

• Resende, Tiago Filipe Coelho

Advisors dc:contributor.advisor

• Soares, Cláudio Manuel
• Rocha, I.

Statement dc:rights

• openAccess

Language dc:language.iso

eng

Handle dc:identifier.uri

https://hdl.handle.net/1822/60059

OAI record oai:identifier

http://repositorium.sdum.uminho.pt/oai/request?verb=GetRecord&metadataPrefix=dim&identifier=oai%3Arepositorium.uminho.pt%3A1822%2F60059