The choice of threshold used for counting the top hits is inevitably, to some extent, arbitrary. In general, the signal to noise ratio is expected to be higher for more significant P-value thresholds but there are progressively fewer hits that achieve more stringent thresholds. Thus, there is a trade off in choosing a threshold that is reasonably stringent but allowing a reasonable number of hits. A threshold of P<10–5 was a benchmark threshold used for reporting signals within the WTCCC study and we adopted this as an appropriate compromise. However, we note that our findings remain unchanged with a higher (P<5×10–6) or lower (P<5×10–5) threshold (data not shown). At these thresholds, RDC schizoaffective disorder, bipolar type again stood out as the only diagnostic subset having a significantly increased number of hits, although as would be expected, the actual number of hits was less or more respectively.
