GenPPI: um software autônomo para predição ab initio de redes de interação entre proteínas bacterianas

Abstract

Protein-protein interactions play a key role in determining the outcome of most cellular processes. Correctly identifying protein interactions and the PPI networks they comprise is of fundamental importance for understanding the molecular mechanisms within the cell. This can provide useful insights in performing critical tasks such as manufacturing drugs and vaccines against diseases caused by infectious agents. Computational approaches are used combining various sources of biological data in order to predict protein interactions with satisfactory levels of reliability. In this work, we propose a new autonomous bioinformatics software (GenPPI) for ab initio prediction of interaction networks between bacterial proteins. The proposed solution analyzes genomes looking for evidence of evolutionary events that indicate protein interactions. Namely, conserved gene neighborhood events, gene fusion and conserved phylogenetic profile. This work also introduces a new heuristic for pairwise comparison of protein amino acid sequences. As a result, we first demonstrate the effectiveness of the proposed heuristic by comparing its accuracy with BLASTp, the main heuristic algorithm for comparing protein sequences. The accuracy of the two heuristic algorithms is estimated by checking which one is closest to the exact Needleman-Wunsh algorithm used for comparing biological sequences. The proposed heuristic surpassed BLASTp, presenting greater accuracy in the pair-by-pair comparison of proteins and shorter processing time. Subsequently, the biological reliability of the computational predictions performed is verified. Therefore, phylogeny analyzes were performed using data generated by the program, after processing genomes of bacterial genera selected as case studies. 28 genomes of the genus Dietzia, 45 of Rhodococcus, 50 of Corynebacterium and 81 of Aeromonas were analyzed. The phylogeny analyzes performed demonstrate correctness and biological reliability for the protein interaction networks predicted by the developed software. Finally, the quality of interaction networks generated by GenPPI is compared with a STRING network, the main state-of-the-art tool of this work. This comparison shows that the proposed solution is capable of generating networks of as good quality as the STRING networks. It is worth mentioning that, with this solution, a deficiency identified in the state of the art, the unavailability of computational tools to predict PPIs without neglecting new proteins, is addressed. The developed software is available for download on the site: <https://genppi.facom.ufu.br/> or on the repository: <https://github.com/santosardr/genppi>, where also contains a user guide.