Comorbidity is an impactful medical problem that is attracting increasing attention in healthcare and biomedical research. However, little is known about the molecular processes leading to the development of a specific disease in patients affected by other conditions. We present a disease interaction network inferred from similarities in patients' molecular profiles, which significantly recapitulates epidemiologically documented comorbidities, providing the basis for their interpretation at a molecular level. Furthermore, expanding on the analysis of subgroups of patients with similar molecular profiles, our approach discovers comorbidity relations not previously described, implicates distinct genes in such relations, and identifies drugs whose side effects are potentially associated to the observed comorbidities.Comorbidity is the tendency for one patient to have an altered risk of developing a second disease when they are already suffering from a specific one. Comorbidity incidence increases with age and has a high impact on life expectancy, which decreases considerably in the presence of a handful of simultaneous diseases 1 as is commonly observed in ageing populations 2 . Additionally, the presence of comorbid conditions has a high economic impact as shown, for example, by the increase of 150% of the cost associated to diabetes for people who are also affected by heart disease 3 . Thus, it is clear that controlling patient-specific risks of future comorbidities could increase life expectancy and reduce public health expenditure 4 .In the research area of comorbidity, tens of disease-disease interaction networks have been published since 2007 5 , using a variety of data types, such as gene expression profiles 6 , combination of disease genes and protein-protein interaction networks 7 , miRNA expression 8 , the microbiome 9 , medical claims 10 , medical records 11 , human symptoms 12 , insurance claims 13 , and mixed information 14 . The Jensen et al. study considers that patients with the same disease might present different risks of developing secondary diseases based on their medical history 11 , which can be a consequence of the existence of different clinical phenotypes within multifaceted conditions as described in chronic obstructive pulmonary disease 15 . Therefore, in this study we set out to explore the molecular bases of comorbidity using patients' transcriptomic profiles to define personalized comorbidity risks.In a previous study based on differential gene expression meta-analyses 16 , we detected that inversely comorbid Central Nervous System disorders and cancers presented significant overlaps between genes deregulated in opposite directions in the two sets of diseases, providing initial molecular evidence for such comorbidity relations. In this new study we have explored this principle at a different level, calculating differential expression profiles for each patient to reduce samples' tissue of origin effect, and defining a patient similarity network including over 6,000 patients affected by 1...