Disease-tissue Associations

The disease-tissue association matrix was generated according to the method defined by Lage et al.²⁷. The association of a tissue and a disease was estimated by measuring their co-occurrence in PubMed abstracts relative to the number of abstracts mentioning the disease or tissue term alone. The disease search terms was limited to PubMed abstracts in which they were qualified as major MeSH²⁸ topics of the article, while the tissue search term was more general limiting to a MeSH topic (no requirement for major). For example, search terms “psoriasis[majr]” and “skin[mh]” was used to measure abstracts that mention psoriasis or skin alone respectively. In order to measure the co-occurrences, the search term “psoriasis[majr] AND skin[mh]” was utilized. The maximal association score (MAS) was computed using the results from the search terms based on the approach using Ochiai’s coefficient²⁹, and then normalized by the sum of all OCs for the each disease as shown previously⁷. These scores range between 0 and 100 with 100 representing the most specific tissue for that particular disease. The entire MAS matrix can be found in Supplementary Table S2. For tissue assignments, we picked the top three tissues with highest MAS score provided they met the threshold score of > = 8%. Increasing this number further did not change the number of observed expression trends significantly (Supplementary Table S3). The final step for assigning a single tissue to each gene-disease association was based on picking the tissue that either showed the highest levels of mRNA/gene expression or could be detected at the protein level (Supplementary Table S1).