Borschbach, M.; Hauke, S.; Pyka, M.; Heider, Dominik:

Opportunities and limitations of a principal component analysis optimized machine learning approach for the identification and classification of cancer involved proteins

In: Communications of SIWN, (2009) ; Nr. 6, S. 85-89
ISSN: 1757-4439
Zeitschriftenaufsatz / Fach: Biologie; Medizin; Informatik
Abstract:
The computational reduction of the input dimension
of different parts of the input layer for the prediction of
functional protein classes is presented. The pro teins are preprocessed
and two feed forward neural networks were trained
with a dataset containing small GTPases and non~GTPases.
The results of the principle component analysis (PCA) optimized
neural network and the neural network without PCA
were compared. The computational complexity and the prediction
accuracy are key aspects of the discussion of the statistical
verified simulation results. Moreover, the mathematical
background of the PCA and therefore the influence of the PCAdimension
on the prediction quality as an underlying optimization
problem based on the necessary selection of eigenvectors
of the feature space is considered.