Forum:Poorly designed "omics" experiment

Question

[Baby Bioinformatician]

Hi Biostars,

Am trying to figure out what can go wrong in the design of omics based studies. Could you please share with me any links to articles in which the authors used an "omics" approach but the study was wrongly designed and hence probably came up with wrong conclusions.

Thanks

[Devon Ryan]

The issue with poorly designed experiments is usually not "wrong conclusions" but instead "no conclusions".

[ATpoint]

In other words, finding NO significant differences between underpowered experiments does not mean there are indeed NO differences but your experiment was not suitable or powered-enough to find them. Typical example would be low (or no) replicates in *-seq experiments, too shallow read depth or insufficient read length (the latter I think mainly for things like structural variant calling, isoform detection...). Replication number is of course a function of money and availability of the biological material (e.g. number of available patients, number of available cells (stem cells, sometimes one has to pool several donors to get enough on the cost of replication)) so underpowered experiments can be a result of circumstances rather than poor planning. Too bad statistics does not care about the reason. Underpowered will remain underpowered. Another thing would be batch effects, e.g. combining different datasets in one statistical analysis. Example would be to have some cancer samples from a public repository, e.g. TCGA, but then take normal controls for e.g. RNA-seq from a completely different publication. No way to distinguish biological from technical (=batch) effects. Same goes for combining samples from paired-end and single-end experiments.

[GenoMax]

omics experiment is a really broad term and you should define what you exactly have on mind. I see that you have only referred to sequencing technologies in post tags.

In biology things happen over time. It may be difficult to design a true omics experiment (where you look at DNA/Protein/Metabolites/Cells/Organs) in one place. Time/effort/cost involved to truly follow these entities (and changes over time) may simply be beyond our current analytical capabilities (and financial constraints).