Tough one. There are several issues to consider. 1) most "omics" data depends on MS abundance and MS is not quantitative across instruments, ionization modes, and labs. Therefore, you need to autoscale every dataset to some common reference standard and deal with the data in a relative abundance domain, not absolute abundance. 2) You will undoubtedly not see the same peaks across labs in any of the data. Even replicates of the same sample back-to-back on the same MS will only give you 20-80% replication depending on your threshold, so if you have no replicates of a peak from a lab, you have to ignore any peaks missing or new in your analysis. You can only compare what is present. Zero (or below detectable limits) doesn't have the same meaning in these data. 3) Now you need to take those RAs that are present in all the data and calculate a coefficients of variation for each individual peak. If you have enough data, then you might check it for normality or log(normality) first, but I've NEVER seen enough replicates to do that with any statistical confidence. 4) Now you need to set a CV or standard deviation threshold that you can use for each replicated peak to say it's either up or down regulated. Again, you could set this by a t-test for outliers, but I've never seen enough replicates to get a t-value at 95% confidence that would produce an outlier.

Unfortunately, "omics" work generally violates the first rule of chemimetrics: you need a minimum of 3 independent samples for every feature that you are measuring in your data. The old company Beyond Genomics hit that wall when they were measuring 300,000 independent features in just 20 disease and 20 control mice. Statistically, they could find nothing that would replicate and it cost them $20M to do the study. They folded when no one would pay to use their platform.