Mixed outcomes for computational predictions

Independent validation of induced overexpression efficiency across 242 experiments shows a success rate of 39

Although numerous studies containing induced gene expression have already been published, independent authentication of their results has not yet been performed. Here, we utilized available transcriptomic data to validate the achieved efficiency in overexpression studies. Microarray data of experiments containing cell lines with induced overexpression in one or more genes were analyzed. All together 342 studies were processed, these include 242 different genes overexpressed in 184 cell lines. The final database includes 4,755 treatment-control sample pairs. Successful gene induction (fold change induction over 1.44) was validated in 39.3% of all genes at p < 0.05. Number of repetitions within a study (p < 0.0001) and type of used vector (p = 0.023) had significant impact on successful overexpression efficacy. In summary, over 60% of studies failed to deliver a reproducible overexpression. To achieve higher efficiency, robust and strict study design with multi-level quality control will be necessary.

The challenges of replication

Interpreting the first results from the Reproducibility Project: Cancer Biology requires a highly nuanced approach.