Proteomics

Mass spectrometry has become a well established method for identifying and characterizing peptides and proteins. In combination with separation techniques like 2d-gel electrophoresis, it can be used to identify individual proteins in samples of tissue or cell cultures. Particularly for medical applications, great hopes are being placed in proteomics, because the protein content of a cell gives more precise information about diseases like cancer (and many others) than its genetic constitution.

Analyzing proteomic data of organisms is significantly more difficult as compared to genetic data, because of their still more complex nature and their large variation (inter-individual, intra-individual as well as temporal). This is where pattern recognition methods show their particular strengths.

Pattern recognition methods may be highly valuable tools for proteomics in many respects:

Intelligent Learning Methods Complement or Replace
Model-based Approaches

Pattern recognition systems are usually based on intelligent learning methods like Artificial Neural Networks or Support Vector Machines. These methods automatically extract and utilize the knowledge, which is implicitly contained in collections of measured data. They do not need any prior information about the mechanisms underlying the data-generating processes. For these reasons, increasing use is being made of pattern recognition methods to develop automatic diagnostic systems (for example for detecting cancer).

Characteristic Features

In addition to the practical value of a trained pattern recognition system when used for analyzing unknown data, these methods provide the possibility of a subsequent reanalysis, by which the decisions made by the adapted system become comprehensible.
One important reanalysis method consists of analyzing those features which have been found by the system to be necessary and useful for decision making. This may lead to the discovery of important and useful biomarkers.

In summary, pattern recognition methods are able to extract knowledge from data and utilize it for answering questions, even if the chemical or biological mechanisms underlying the given problem are not known. Conversely, the information about which of the many features have been found useful for solving the task might provide valuable insights into the nature of these mechanisms.

Combination with Genetic Data

Pattern recognition methods allow to analyze high-dimensional data, which may be combined of several parts coming from different sources. Particularly, the proteomic information contained in mass spectra may be combined with genetic data to form joint data records. This may, for example, help to find a multi-factorial genetic basis of a certain disease. Our software PATTERN EXPERT amspect allows to supplement the mass spectra with additional information like genetic data and to perform a combined analysis.

Combination with Microarray Data

As described above, PATTERN EXPERT amspect is able to analyze mass spectra in combination with additional user supplied information. In particular, this can be a very useful tool if microarray data are added.