An artificial neural network model for predicting the subcellular localization of photosensitisers for photodynamic therapy of solid tumours.

Photodynamic therapy (PDT) is a promising modality for the treatment of tumours based on the combined action of a photosensitiser (PS), visible light and molecular oxygen, which generates a local oxidative damage that leads to cell death. The site where the primary photodynamic effect takes place depends on the subcellular localization of the PS and affects the mode of action and efficacy of PDT. It is therefore of prime interest to develop structure-subcellular localization prediction models for a PS from its molecular structure and physicochemical properties. Here we describe such a prediction method for the localization of macrocyclic PSs into cell organelles based on a wide set of physicochemical properties and processed through an artificial neural network (ANN). 128 2D-molecular descriptors related to lipophilicity/hydrophilicity, charge and structural features were calculated, then reduced to 76 by using Pearson's correlation coefficient, and finally to 5 using Guyon and Elisseeff's algorithm. The localization of 61 PSs was compiled from literature and distributed into 3 possible cell structures (mitochondria, lysosomes and "other organelles"). A non-linear ANN algorithm was used to process the information as a decision tree in order to solve PS-organelle assignment: first to identify PSs with mitochondrial and/or lysosomal localization from the rest, and to classify them in a second stage. This sequential ANN classification method has permitted to distinguish PSs located into two of the most important cell targets: lysosomes and mitochondria. The absence of false negatives in this assignation, combined with the rate of success in predicting PS localization in these organelles, permits the use of this ANN method to perform virtual screenings of drug candidates for PDT.

Comparative study of neural networks and least mean square algorithm applied to the optimization of cosmetic formulations.

In this work, a comparative study between two methods to acquire relevant information about a cosmetic formulation has been carried out. A Design of Experiments (DOE) has been applied in two stages to a capillary cosmetic cream: first, a Plackett-Burman (PB) design has been used to reduce the number of variables to be studied; second, a complete factorial design has been implemented. With the experimental data collected from the DOE, a Least Mean Square (LMS) algorithm and Artificial Neural Networks (ANN) have been utilized to obtain an equation (or model) that could explain cream viscosity. Calculations have shown that ANN are the best prediction method to fit a model to experimental data, within the interval of concentrations defined by the whole set of experiments.