Technical strategies to reduce the amount of "false significant" results in quantitative proteomics

When the p-value is set at <0.05 in statistical group comparisons, a 5% rate of "false significant" results is expected. In order to test the reliability of our 2-DE method, we loaded each of 24 gels with equal-sized samples (200 mu g protein from pooled rat brain, pH 4-7, stained with ruthenium fluorescent stain for visualization) and statistically compared the first 12 gels with the last 12. In numerous experiments the rate of significant differences found far exceeded 5%. Several factors were identified as causing the following rates of false significant differences in spot intensities: (i) running samples in two different 2-DE runs (42%), (ii) running second dimension gels produced in two different gel casters (16%), (iii) normalizing the entire gel instead of separately normalizing several different gel zones (11%), (iv) using IPG strips from different packages (19%), (v) dividing the whole sample into subgroups during software analysis (9%). After controlling for all these factors, the rates of "false positive" results in our experiments were regularly reduced to approximately 5%. This is an indispensable prerequisite for avoiding too high a rate of false positive results in experiments in which different subgroups are compared statistically.