Pattern Recognition Letters
Genetic-Based EM Algorithm for Learning Gaussian Mixture Models
IEEE Transactions on Pattern Analysis and Machine Intelligence
Bioinformatics
Hi-index | 0.00 |
Nuclear Magnetic Resonance spectroscopy is a powerful technique for the determination of protein structures and has been supported by computers for decades. One important step during this process is the identification of resonances in the data. However, due to noise, overlapping effects and artifacts occuring during the measurements, many algorithms fail to identify resonances correctly. In this paper, we present a novel interpretation of the data as a sample drawn from a mixture of bivariate Gaussian distributions. Therefore, the identification of resonances can be reduced to a Gaussian mixture decomposition problem which is solved with the help of the Expectation-Maximization algorithm. A program in the Java programming language that exploits an implementation of this algorithm is described and tested on experimental data. Our results indicate that this approach offers valuable information such as an objective measure on the likelihood of the identified resonances.