First I checked for duplicates in the 1 to 1 human to anole ortholog table. There were two:
RPP14 is listed twice with two different human gene ids. Both have the same start but different endings. The anole orthologs have different starts but are close and fall in order in the gene list. Since these two entries would be treated at a block of two genes, I removed the second entry from the table before doing the permutation analysis.
Human loci: chr3:58291974-58305816, chr3:58291974-58305920 Anole loci: chr2:187318639-187323381, chr2:187318070-187318486
LEKR1 is similarly listed twice with two human gene ids. The loci are different but close and fall in order in the gene list, so I removed the second entry.
Human loci: Chr3:156543270-156697347, chr3:156544096-156763918 Anole loci: Chr3:15191770-15221921, Chr3:15126986-15152743
Next I generated 1,000 lists of unique genes with the same chromosomal distributions and block sizes as were observed in the genes missing in the bird genome. Each iteration does the following:
For each block of genes size n in the missing gene set I randomly choose a seed gene on the given genome. I check that the n genes following this gene are on the same genome in both human and anole and that none of the genes are already on the list of genes for that permutation. This continues until I get a full set of 274 randomly selected genes.
Using this list of 1,000 gene sets, I counted the number of OMIM ids associated with each gene and the number of OMIM disease terms associated with each set of genes. The genes missing in the bird genome do not have a significantly different number of OMIM ids associated with them (two-sided permutation test p-value: 0.42). However, none of the permutations had a number of OMIM disease terms as extreme as was observed for the genes missing in the bird genome. Therefore a two-sided permutation test p-value is less than .001 indicating that the genes missing in the bird genome have a significantly fewer OMIM disease terms than would be expect by chance when accounting for the chromosomal distribution and block sizes.
Lastly I looked at the number of genes associated with each mouse phenotype in both the permutation gene sets and the set of genes missing in birds.
For each phenotype I compared the number of genes associated with that phenotype in the missing gene set to the distribution of the number of genes associated with that phenotype in the permutation gene sets. I report the number of associated genes in the missing set, the mean and standard deviation for the number of associated genes in the permutation gene sets and a two-sided permutation p-value. This value is calculated as above, namely the number of permutations finding a count as extreme as that observed in the missing gene set is divided by the number of permutations. These values are in the file: Mouse_pheno_perm.txt.
I tried several approaches for multiple test correction of the above 7,874 phenotype permutation tests. For a given phenotype, if we never observed a gene count as extreme as that seen in the missing gene set then the calculated p-value is 0/1000 = 0. However since we can't know whether such an extreme value would have occurred with more permutations, for multiple test correction these values were set to 0.001, an upper bound for the true p-value.
A Benjamini-Hochberg false discovery rate (FDR) adjustment on all 7,874 tests showed no significant phenotypes at the 0.05 level.
If we assume that phenotypes with p-values set to 0.001 above would behave similarly with 10,000 permutations and we set those p-values to 0.0001, FDR correction shows 47 phenotypes with q-values < 0.05
Alternatively we can restrict our examination to only those phenotypes with larger effect sizes. If we only look at phenotypes for which the number of expected and observed genes differs by more than one, our number of tests is reduced to 376. With 1,000 permutations we find 11 significant FDR q-values at a 0.05 level. All of these are phenotypes where the number of missing genes is greater than one and the mean number of missing genes in permutations is less than one. The q-values resulting from this approach are given in the file Mouse_pheno_perm_q.txt.