SELDI MS data analysis in cancer proteomics

At Karolinska Biomics Center (KBC), Karolinska Institutet, we are working with genomics, proteomics, metabonomics and bioinformatics in close collaboration with clinical activity. This gives us an exceptional opportunity to work with tasks highly relevant for the treatment of patients, mainly with cancer. Further, the research fields of clinical proteomics and bioinformatics are closely associated at KBC. For example, we are working a lot with understanding, interpretation, and improvement of proteomic mass spectrometry data analysis.

Surface-enhanced laser desorption ionization mass spectrometry (SELDI MS) is a chip based method for protein profiling. The sample is applied to a chemical surface, which selectively binds proteins. The bound proteins are then analysed by time-of-flight (TOF) mass spectrometry. Spectra from different samples are compared in order to find proteins characterising some special phenomenon in a group of patients or cells. As an example, we are searching for proteins characterising good or bad response to a medical treatment in samples from a group of patients. As a result, we wish to be able to predict a response to therapy and further hopefully enable personalised medicine.

A lot of noise and systematic variations are included in the MS data from such a study. Therefore, it is of the highest importance to be able to understand and optimise the processing of this kind of data to get the maximum and true information out from it. This includes standardisation between spectra (normalisation), other kinds of pre-processing and transformation of data, and in the end pattern recognition by supervised or unsupervised data analysis methods such as principal component analysis (PCA) and many more.

From this background, we hereby suggest three undergraduate student projects. We are looking for co-workers with experiences in numerical analysis, chemometrics, analytical chemistry, bioinformatics and/or programming. Familiarity with R and or MATLAB is an advantage. Also, You should be able to work independently to some extent.


1. Standardisation/normalisation of SELDI proteomic data
Main objective:
How to standardise/normalise SELDI data in a way to get an as true protein concentration as possible from the data analysis?
Interim targets:
Find, implement and evaluate methods for standardisation/normalisation relevant for SELDI data. E.g.:
• Lowess (locally weighted linear regression) normalisation
• Quantile normalisation
• Normalisation to equal area (TIC)
• Normalisation to median intensity

2. Preprocessing of SELDI proteomic data
Main objective:
How to preprocess SELDI data in a way to get an as true protein concentration as possible?
Interim targets:
Find, implement and evaluate methods for SELDI data preprocessing. E.g.:
• 2:nd derivative
• MSC (multiplicative scatter correction)
• wavelets

3. Alignment of SELDI proteomic data
Main objective: How to align SELDI data in a way to get an as true protein peak agreement as possible?
Interim targets:
Several different methods for alignment have been published during the last decade.
Find, implement, evaluate and compare methods for alignment of SELDI data. E.g.:
• ProSpect
• PCA based methods
• Bucketing
• Segment-wise alignment


General Interim targets for all the projects are:

I. A search for literature to find information related to the project.

II. Determination how to evaluate the methods?
• Analysis of data from a repeatability study – some limited amount of data is available.
• Run a standard protein mixture with known concentrations on the SELDI, and analyse this data regarding: how true is the relationships between peaks? repeatability? and more.
• Compare with the Ciphergen protein chip software (standard method).

III. Implement the methods for data preprocessing and evaluation (in R or MATLAB) for the numerical analysis.