Desarrollo de modelos de clasificación para predecir la mutagenicidad de moléculas orgánicas
Fecha
2016-06
Autores
Huici Corrales, Adonis Jorge
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Central “Marta Abreu” de Las Villas. Facultad de Química Farmacia. Departamento de Licenciatura en Química
Resumen
En el proceso de desarrollo de fármacos, los ensayos de toxicidad son necesarios para preservar la salud de los pacientes que consumen el producto. Dentro de estos ensayos, los de mutagenicidad resultan especialmente costosos y difíciles de realizar. Una forma de reducir el número de estos ensayos es el uso de técnicas auxiliares como el cribado in vitro e in silico de sustancias orgánicas con potencialidades farmacológicas.
El presente trabajo constituye un estudio in silico preliminar para desarrollar modelos de clasificación con alto poder predictivo de la mutagenicidad de moléculas orgánicas. Para ello se confeccionaron 4 Bases de Datos de sustancias con mutagenicidad reportada de acuerdo a 4 ensayos biológicos; cuyos casos fueron etiquetados como mutagénicos y no mutagénicos.
Las bases de datos fueron curadas y posteriormente utilizadas para el cálculo de descriptores moleculares 0D-2D, implementados en los programas ISIDA_Fragmentor y DRAGON. Los descriptores más representativos junto a la variable respuesta, se emplearon para entrenar varios modelos de clasificación, que fueron validados interna y externamente. Los de mejor desempeño resultaron ser los árboles de decisión Random Forest que usan los descriptores del DRAGON. Estos alcanzaron un 75% de predicción de la serie externa para la Base de Datos Bacterias, lo que es aceptable si se considera la heterogeneidad estructural de las bases de datos.
In drug design, the toxicity assays are necessary to preserve the health of the patients that consumes the product.Within this assays, mutagenicity ones are specially expensive and hard to perform. A way to reduce the number of this assays is applying auxiliary techniques like the in vitro and in silico screening of organic substances with pharmacological potentialities. This work is a preliminary in silico study to develop classification models with high predictive power of organic molecules mutagenicity. For this purpose 4 Databases of substances with reported mutagenicity were prepared according 4 biological assays; whose cases were labeled as mutagenic and non-mutagenic. The databases were cured and then used for calculate 0D-2D molecular descriptors, implemented in ISIDA_Fragmentor and DRAGON softwares. Most representatives descriptors with the response variable were used for training of several classification, that were validated internally an externally. Best performance models were the decision trees Random Forest using the DRAGON descriptors. This models reach the 75% of external dataset prediction in Bacterias dataset, which is acceptable if the structural heterogeneity of databases is considered.
In drug design, the toxicity assays are necessary to preserve the health of the patients that consumes the product.Within this assays, mutagenicity ones are specially expensive and hard to perform. A way to reduce the number of this assays is applying auxiliary techniques like the in vitro and in silico screening of organic substances with pharmacological potentialities. This work is a preliminary in silico study to develop classification models with high predictive power of organic molecules mutagenicity. For this purpose 4 Databases of substances with reported mutagenicity were prepared according 4 biological assays; whose cases were labeled as mutagenic and non-mutagenic. The databases were cured and then used for calculate 0D-2D molecular descriptors, implemented in ISIDA_Fragmentor and DRAGON softwares. Most representatives descriptors with the response variable were used for training of several classification, that were validated internally an externally. Best performance models were the decision trees Random Forest using the DRAGON descriptors. This models reach the 75% of external dataset prediction in Bacterias dataset, which is acceptable if the structural heterogeneity of databases is considered.
Descripción
Palabras clave
ADN, Sustancias Mutagénicas, Modelación (Q)SAR, Industria Farmacéutica, Descriptores Moleculares