Implementação de árvores de decisão para propriedades tridimensionais em linguagem Python

Abstract

Bioinformatics is a field of research that makes use of computational techniques for bioilogical results. Rational Drug Design is an importante field of research focusing on the interaction between macromolecules, called as receptors, and small molecules, called as ligands. The objective is to investigate the best fit between these molecules to perform or inhibit specific functions, which can be measured by the estimated Free Energy of Binding (FEB). Most studies consider the receptor as a rigid structure, however a more realistic method must consider not only the ligand as a flexible structure, but also the receptor. This flexibility can be simulated by means of a technique called Molecular Dynamics (MD) simulation. After an in silico RDD experiments, the most promissing ligands for a particular receptor are tested in vitro and a new drug may be created. The algorithm implemented in this work induces a regression-tree considering the three-dimensional properties of the receptor’s atoms the leads to a good estimated FEB value. This algorith makes use of the coordinates to split a node into two parts, where the atom is evaluated in terms of its pose in a block - which represents the best position in the space. A domain expert may then select promising conformations of the receptor from the induced model. This work deals with the implementation of the 3D-Tri algorithm in Python. This language was chosen because of its extensive use in data mining context, its large number of available libraries and the facility of reading and writing in that language. Tests were performed with data of NUNL2 ligand, known efflux pump inhibitor AcrB (Acriflavine resistance protein B). These tests resulted in a model easily interpreted by a domain expert, a characteristic of the 3D-Tri algorithm. This implementation aims to add improvements to the algorithm and extend the use of it, to thus contribute to the RDD process.