The Unresolved problems of peptide drug metabolism : Software-aided approach designed to analyze and predict cleavage sites for natural and synthetic peptides

Abstract

The work of this thesis aims to develop of a new workflow to predict potential Sites of Cleavage (SoCs) in new candidate peptide drugs. The main goal of this workflow is to understand the protease specificity rules from data coming from different sources and with no limitations on peptides structure. To complete this goal, we implemented a new algorithm to store the information from experimental and external sources in a chemically aware database WebMetabase. This database can be further enriched with new data without limitations. Moreover, we were able to process information about cyclic peptides. Than we developed and applied a frequency analysis to reveal the metabolically labile amide bonds defined by similar properties of residues around the SoC towards the specific proteases. Finally, we built several predictive models using a training dataset, where each SoC was described by its molecular descriptors and frequency. Prediction ability of these models was not limited to only natural amino acids. We demonstrated that models have a comparable predictive performance as other public tools such as PROSPERous.

El objetivo de esta tesis doctoral es el desarrollo de uno nuevo sistema para predecir el sitio de hidrolisis (SoC) para péptidos con actividad terapéutica. El objetivo principal de la metodología es el establecer reglas de especificidad catalíticas paras proteasas sin limitaciones en cuanto a fuente de datos y/o estructura. Con este fin, se implementó un nuevo algoritmo para guardar la información en una base de datos que considera la estructura química, WebMetabase. Esta base de datos puede enriquecerse con nueva información sin límites, pudiendo procesar péptidos cíclicos. Se ha desarrollado un nuevo análisis de frecuencias con el fin de encontrar los enlaces amida metabólicamente lábiles, definidos mediante la similitud de los residuos cercanos al SoC. Se construyeron modelos de predicción donde cada SoC se describe con descriptores moleculares y frecuencia. Estos modelos no se limitan a los aminoácidos naturales y son comparables a los obtenidos con herramientas públicas como PROSPERous.