Finding Relevant Parameters for the Thin-film Photovoltaic Cells Production Process with the Application of Data Mining Methods.

A data mining approach is proposed as a useful tool for the control parameters analysis of the 3-stage CIGSe photovoltaic cell production process, in order to find variables that are the most relevant for cell electric parameters and efficiency. The analysed data set consists of stage duration times, heater power values as well as temperatures for the element sources and the substrate - there are 14 variables per sample in total. The most relevant variables of the process have been found based on the so-called random forest analysis with the application of the Boruta algorithm. 118 CIGSe samples, prepared at Institut des Matériaux Jean Rouxel, were analysed. The results are close to experimental knowledge on the CIGSe cells production process. They bring new evidence to production parameters of new cells and further research.

Gene expression profile as a prognostic factor in high-grade gliomas.

Some clinical factors have been useful in predicting prognosis in high-grade gliomas, however, unexpected differences in survival time have generated attempts to search for more precise parameters. It is clear that tumour behaviour depends mostly on gene alterations. Known single gene alterations failed to accurately define survival time, however, recently, the gene profiling based on microarray technology has raised hopes. Our aim was to assess whether the genetic predictor exceeds clinical parameters in the prognosis of malignant gliomas. We performed gene expression analysis of 28 gliomas (3 grade II, 10 grade III and 15 grade IV, according to WHO classification), and 5 control, normal brain samples, using Clontech oligonucleotide arrays with 3,757 known genes. The signal-to-noise statistics was used to separate classes, and the leave-one-out method was used to assess the smallest number of genes make it clear with a minimal cross-validation error. All gliomas, or only high-grade tumours, were clearly separated from the normal brain samples using 7 or 9 most differentially expressed genes. Hierarchical clustering failed, but the fuzzy c-means method was useful in high-grade gliomas to find a gene prediction model, which, with clinical factors, was assessed in survival analysis. Univariate analysis demonstrated that age, WHO grade (IV vs. III), radiation dose (> or = 50 Gy vs. 42 Gy), postoperative KPS score (100 points vs. others), neurological deficit as the first sign of the disease vs. others, and gene expression profile were significant predictors of survival. In multivariate analysis, the gene expression profile remained the only independent predictor (p = 0.007). Thus, our conclusion is that gene expression pattern predicts outcome in high-grade gliomas independently of other factors.