Desarrollo de un predictor de interacciones entre dominios de proteínas basado en nuevos descriptores numéricos de secuencias de aminoácidos
Fecha
2016-11-15
Autores
Romero Molina, Sandra
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Central “Marta Abreu” de Las Villas. Facultad de Matemática, Física y Computación. Departamento de Ciencias de la Computación
Resumen
Las proteínas intervienen en todos los procesos biológicos que ocurren con y entre las células. Particularmente las interacciones entre proteínas determinan muchas de sus funciones, como la acción enzimática en procesos metabólicos, la transmisión de señales y el transporte de sustancias. Por ello, el desarrollo de métodos computacionales de predicción de Interacciones Proteína-Proteína (PPI del inglés Protein-Protein Interactions) resulta uno de los objetivos más significativos de la Biología Computacional moderna. Los mejores métodos de predicción de PPI emplean el análisis de homología entre proteínas, mediante técnicas de alineamiento de secuencias de aminoácidos. Esta estrategia, unido al deficiente uso de casos de pares sin interacción, lleva a que los métodos muestren baja precisión, lo que limita su aplicabilidad para el diseño de proteínas/péptidos.
El presente trabajo describe los resultados obtenidos en la modelación de interacciones dominio-dominio (DDI) empleando nuevos descriptores numéricos libres de alineamiento calculados con el software ProtDCal. Se conforma una base de casos de pares de dominios de proteínas con interacción o ausencia de interacción; obtenidos de las bases de datos 3DID, iPFam y Negatome, en donde cada par de secuencias es representado por un conjunto de descriptores numéricos. Se realiza un pre-procesamiento del conjunto de entrenamiento y se construyen modelos empleando la técnica de aprendizaje automatizado Máquinas de Vectores de Soporte. Los resultados se comparan con métodos externos; mostrando un desempeño superior particularmente en la exactitud de las predicciones.
Proteins are involved in all biological processes occurring within and between cells. Particularly protein interactions determine many of its functions such as enzyme action in metabolic processes, signal transmission and transport of substances. That is why the development of computational methods for predicting protein-protein interactions is one of the most important goals of modern Computational Biology. Best PPI prediction methods used to date analyze homology between proteins by amino acid sequences alignment techniques. This homology dependent strategy, coupled with poor use of case negative information leads to methods show low precision, which limits its applicability to design proteins / peptides. This paper describes the results of the modeling domain-domain interactions (DDI) using new free alignment numerical descriptors calculated with ProtDCal software. An instances base is formed by pairs of protein domains with interaction or absence of interaction obtained from the 3DID, IPFam and Negatome databases; in which each pair of sequences is represented by a set of numerical descriptors. A pre-processing of the training set is performed and models are constructed using the machine learning technique Support Vector Machines. The results are compared with those shown by different external methods; showing superior performance particularly in the accuracy of predictions.
Proteins are involved in all biological processes occurring within and between cells. Particularly protein interactions determine many of its functions such as enzyme action in metabolic processes, signal transmission and transport of substances. That is why the development of computational methods for predicting protein-protein interactions is one of the most important goals of modern Computational Biology. Best PPI prediction methods used to date analyze homology between proteins by amino acid sequences alignment techniques. This homology dependent strategy, coupled with poor use of case negative information leads to methods show low precision, which limits its applicability to design proteins / peptides. This paper describes the results of the modeling domain-domain interactions (DDI) using new free alignment numerical descriptors calculated with ProtDCal software. An instances base is formed by pairs of protein domains with interaction or absence of interaction obtained from the 3DID, IPFam and Negatome databases; in which each pair of sequences is represented by a set of numerical descriptors. A pre-processing of the training set is performed and models are constructed using the machine learning technique Support Vector Machines. The results are compared with those shown by different external methods; showing superior performance particularly in the accuracy of predictions.
Descripción
Palabras clave
Proteínas, Técnicas Wrapper, Dominios de Proteínas