Table 1.

The embedding spaces of the most prevalent cancers (breast, prostate, lung, and colorectal cancer) and their control tissues (breast glandular cells, prostate glandular cells, lung pneumocytes, and colorectal glandular cells) are functionally organized.

Embedding	Intra	Inter	Fold	P-value
Control breast	0.22	0.17	1.29	$2.12\times {10}^{-6}$
Cancer breast	0.23	0.16	1.43	$2.68\times {10}^{-5}$
Control prostate	0.24	0.17	1.41	$2.24\times {10}^{-6}$
Cancer prostate	0.21	0.15	1.40	$1.04\times {10}^{-6}$
Control colon	0.19	0.16	1.18	$4.04\times {10}^{-3}$
Cancer colon	0.21	0.16	1.31	$1.68\times {10}^{-5}$
Control lung	0.19	0.17	1.11	$2.17\times {10}^{-4}$
Cancer lung	0.22	0.15	1.46	$5.32\times {10}^{-6}$
Random example	0.17	0.17	1.00	0.14

The first column, ‘Embedding’, lists the tissues. The second column, ‘Intra’, shows the average Lin’s semantic similarity of those annotations whose embedding vectors cluster together based on their cosine distances in the embedding space. The third column, ‘Inter’, shows the average Lin’s semantic similarity of those annotations whose embedding vectors do not cluster together based on their cosine distances in the embedding space. The fourth column, ‘Fold’, displays how many times the average Lin’s semantic similarity of those annotations whose embedding vectors cluster together based on their cosine distances in the embedding space is higher than of those annotations whose embedding vectors do not cluster together. The fifth column, ‘P-value’, shows the P-value from a one-sided Mann–Whitney U test comparing the Lin’s semantic similarity between annotations whose embedding vectors cluster together and those with non-clustered embedding vectors. This table also includes an example of a randomly rewired PPI network (Random Example). The complete information with all the random tissue-specific PPI networks can be found in Supplementary Table 6.