1. Compute the average fold change and standard deviation for all genes in the comparison.
2. Compute a normalized p-value (NPV) for each gene in the comparison, based on its fold change (fc) value, and the above average (ave) and standard deviation (std) values. Use a normal distribution function (CDF):
if (fc > 0)
NPV = 1.0 - CDF(fc, ave, std)
else
NPV = 1.0 - CDF(-fc, ave, std)
if NPV < 1.0e-15, set to 1.0e-15
3. For each gene in the gene signature, list, or pathway, search against experimental comparisons and extract those comparisons where the gene’s normalized p-value is less than 0.05. This returns a comparison list.
4. Iterate through the comparison list. For each comparison (C), add the normalized p-value to one of two arrays of sums (pv_sum), as follows:
□ For gene signatures and gene lists, add the gene’s normalized p-value to:
● pv_sum(C, up) if the gene’s fold change in the signature (svc) and in the comparison (cfc) are co-regulated.
● pv_sum(C, down) if the gene’s fold change in the signature and in the comparison are anti-regulated.
□ For pathways, add the gene’s normalized p-value to:
● pv_sum(C, up) if the gene’s comparison fold change (cfc) is up-regulated.
● pv_sum(C, down) if the gene’s comparison fold change is down-regulated.
Also, use the logarithm of the normalized p-value to make the final TEA score more human readable:
if (gene_signature OR gene_list)
if ( (sfc > 0 AND cfc > 0) OR (sfc < 0 AND cfc < 0) )
pv_sum(C, up) += -Math.log(NPV)
pv_count(C, up)++
else
pv_sum(C, down) += -Math.log(NPV)
pv_count(C, down)++
if (gene_pathway)
if (cfc > 0)
pv_sum(C, up) += -Math.log(NPV)
pv_count(C, up)++
else
pv_sum(C, down) += -Math.log(NPV)
pv_count(C, down)++
5. Compute the min-LogP average (pv_ave) for each sum:
pv_ave(C, up) = Math.exp(-pv_sum(C,up) / pv_count(C,up) )
pv_ave(C, down) = Math.exp(-pv_sum(C,down) / pv_count(C,down) )
6. Compute a TEA score (pv_tea) for each min-LogP average through a binomial distribution function:
pv_tea(C, up) = 1.0 - Binom( N, pv_count(C,up), pv_ave(C,up) )
pv_tea(C, down) = 1.0 - Binom( N, pv_count(C,down), pv_ave(C,down) )