Microarray gene expression studies are mainly used in two types of experiments: static and time-series. In fact, in the public domain, time-series experiments account for about 40 percent of the datasets. While time-series experiments are widely used to study biological systems, determining the quality of the results can be a fundamental problem according to Ziv Bar-Joseph, assistant professor of computer science and biological sciences at Carnegie Mellon University. Bar-Joseph is the inventor of a computational method that identifies genes missed by current analysis methods.
Some of the problems involved in the analysis of microarray data include determining the number of replicates for each time point, normalizing across different microarray platforms, and dealing with noise artifacts.
"Even if all these artifacts and noise are perfectly corrected, the resulting time-series profile may not really represent the underlying expression profile," says Bar-Joseph.
Why? One major issue is the sampling rate. In time-series experiments, measurements are taken at different intervals. "If a gene goes up and down in between these intervals, it can be missed even if everything is normalized. You might miss a collection of genes involved in some physiological process or drug response. This is a major issue. This is an error that cannot be caught now even with the most sophisticated normalization tools," says Bar-Joseph.
In addition, different arrest methods used to synchronize cells can cause many genes that are not involved in the system under investigation to react in response to the arrest-and-release treatments. This makes it hard to distinguish between expression profiles that are the result of genes participating in the system and changes stemming from the treatment itself. "The measurements are real in the sense that the genes are going up, but not as a result of the process you're studying, but rather because of the treatment," says Bar-Joseph.
The method Bar-Joseph pioneered aims to solve these problems and see if 'what you see is what you get.' Is what you observe in an expression profile really the way the gene operates in the system?
.
The experimental protocol involves only one additional microarray experiment in which a mixture of cells is profiled throughout the process. The averages/values obtained from this experiment are compared with the time-series profiles. If they agree, one can claim that what you see is what you get. If the time-series profile is higher than the average, one can conclude that the experiment results are a reaction to the treatment or something else unrelated to the process. If the average is high, but the time-series is low, then one can say you missed that gene, perhaps due to sampling problems.
"We did follow-up experiments and chose genes that could not have been identified based on time-series experiments alone, and the method picked them up correctly," says Bar-Joseph.
The program can be downloaded for free at:
http://www.cs.cmu.edu/~zivbj/checksum/checksum.html