abstract
- © Springer International Publishing AG 2017. Molecular Docking faces problems related to Curse of dimensionality, due to the fact that it analyzes data with high dimensionality and few samples. (Ligand-Based Virtual Screening) conducts studies of docking among molecules using common attributes registered in data bases. This branch of Molecular Docking, uses Optimization methods and Machine learning algorithms in order to discover molecules similar to known drugs and can be proposed as drug candidates. Such algorithms are affected by effects of Curse of dimensionality. It this paper we propose to use LMC complexity measure (Complexity of Lopez-Ruiz, Mancini, and Calbet) [1] as similarity measurement among vectors in order to discover the best molecules to be drugs; and present an algorithm, which evaluates the similarity among vectors using this concept. The results suggest that application of this concept on Drug Example vectors; in order to classify other vectors as drugs candidates which is more informative than individually searching for vectors. Since the Drug Examples show a global similarity degree with drug candidate vectors. The aforementioned similarity degree makes it possible to deduce which elements of the Drug Examples show higher degree of similarity with drug candidates. Searching of vectors through individual comparison with Drug Examples was less efficient, because their classification is affected by the Drug Examples with a higher number of global discrepancies. Finally, the proposed algorithm avoids some of the Curse of dimensionality effects by using a ranking process where the best drug candidate vectors are those with the lowest complexity.