A second study

[60] applied four different pathway analyses methods and detected 17 pathways that were significantly enriched for association with MDD. Their top pathways included long-term depression, calcium signaling, arrhythmogenic see more right ventricular cardiomyopathy, and cell adhesion molecules. Song and Lee [61] implicated a central role of negative regulation of transcription and nucleic acid metabolism in MDD. These reports lack the indications of coherence seen in SCZ and BD, and suggest that larger sample sizes are required. At present, the published sample sizes of genomic studies of ADHD are limited, with samples sizes of each of the four studies we identified at or below 1000 cases 39, 62, 63 and 64]. The largest study [63] suggests the involvement of synaptic mechanisms

in ADHD; however there is no convergence on pathways across the different studies. Fortunately, genomics studies of ADHD are in progress that should markedly increase the available sample sizes. As a consequence of this, pathway analyses should become more informative. We reviewed 42 studies that reported on biological pathways involved in five major psychiatric disorders. For SCZ and BD, where studies were based on sizable samples, pathway results tend to converge. For ASD, and especially MDD and ADHD, there was much less convergence suggesting that current pathway studies for these disorders are underpowered. The importance of sample size and statistical power cannot be overemphasized. An illustrative power analysis shows that to detect an effect size of VX-765 a gene-set equivalent to a genotypic relative risk of 1.1 requires ∼23,000 cases and 23,000 controls (assuming 80% power and a self-contained test in 500 gene-sets) [65]. At present, only SCZ has attained such numbers. Another important issue is that gene-set definitions vary considerably across studies. Gene-sets are derived from public databases (e.g., KEGG, GO, or Reactome) or are based on expert curation. Gene-sets available in public databases are neither complete, error-free, nor unbiased 66, 67 and 68].

To illustrate, we evaluated 1027 genes that were annotated by experts as being present in the synaptic compartment, and verified by repeated lab experiments to be active in the synapse 27 and 69]. Most (58%) of these genes had no known pathway Interleukin-2 receptor in the KEGG database. The GO database contains 22 unique terms in the component category, containing ‘synaptic’ or ‘synapse’ for a total of 540 unique genes. Of the 1027 expert-curated synaptic genes, only 225 (22%) are annotated as being present at the synapse in GO. Of 540 GO synaptic genes, 315 have not been experimentally validated as playing a role in the synapse. This single (albeit important) example may or may not generalize; however, bias or unreliability in public databases is certainly possible, and may be a critical limitation for pathway analyses.