Phenotypes are the result of the expression of specific genetic backgrounds submitted to the influence of changing environmental conditions 
. Thus, both the development and resulting symptoms of a given pathology are conditioned by interacting elements at multiple interconnected levels (from molecular to social levels) 
. These complex interactions can be represented as networks to be analyzed using the principles of Network Theory 
. In this sense, Network Medicine emerged as a new field to study the relationships among diseases and disease-causing genes 
. Generally, data from genetic association studies establish the basic information for these analyses. Most of these data are available from different public repositories, for instance, Online Mendelian Inheritance in Man (OMIM) 
and Orphanet 
. This information can be projected onto networks also known as diseasomes (i.e. “the human disease network” and “the orphan disease networks”) 
. These diseasomes open the possibility to work on different types of network projections, treating networks as graphs, which can be used to detect emergent information. For instance, disease-to-gene associations represent bipartite edges (two different types of nodes in every edge) and conform a bipartite graph (as shown in the schematic representation in ). On the other hand, projections of gene-to-gene edges and disease-to-disease edges can be inferred from the initial bipartite graph as two different “unipartite” graphs (each with only one type of node). Hence, edges in both inferred unipartite graphs represent either genes associated by a same disease () or diseases associated through a same gene (these edges were not considered in this study), respectively. The first type of projections (gene-to-gene) are disease-causing gene networks and the second ones (disease-to-disease edges) are generally known as disease networks 
. Network-based methods enable us to find disease modules that may be understood as all molecular relationships involving disease-causing genes and other genes related to the same pathological processes 
. In fact, several different biomolecular interactomes based on physical, metabolic or functional interactions have been used to capture some frames of the biological complexity associated with pathologies 
. In this case, one of the most direct applications of network medicine approaches lies in the systematic exploration of the molecular mechanism shared by “apparently” distinct diseases 
. The emergence of relationships among genes and diseases contribute to obtain more holistic views of the disease origin and environment, to predict new disease-causing genes 
, and possibly to locate new targets for disease diagnosis and/or intervention. All these challenges take part in a wider emergent discipline known as Systems Medicine 
Schematic representation of distinct disease-to-gene relationships.
However, current pathognomonic classifications are influenced by the traditional clinical procedures used during the 19th
century following Osleŕs principles 
. These traditional procedures often tend to overvalue the most evident manifested abnormalities (pathophenotypes), causing a direct impact on how pathophenotypic profiles of patients are registered in the clinic 
. Although it could help the diagnosis, many others pathophenotypes will go unnoticed. As a consequence, most genetic diseases are described as conceptual entities, pathologies, with certain specific clinical features. The disregard of pathophenotypes implies a considerable technical problem for network medicine based methods, since they can be primary consequences of the genetic disturbances. At present, to solve this problem standard phenotypic platforms are required to explore the underlying molecular and cellular mechanisms related to genetic predisposition in developing diseases 
. Nevertheless, some previous works have claimed that the systematic phenotyping procedure requires ontologies to improve biomedical insights on functional gene communities 
. In this case, the use of ontologies can be an interesting advance in the biomedical integration of this information. The Human Phenotype Ontology (HPO) represents a formalization of the semantic relationships 
among different clinical features described in OMIM (abbreviations used throughout the manuscript are reported in ). Although HPO was initially developed to study the phenotypic associations in order to achieve a potential diagnostic use 
, this standardized biomedical knowledge on human abnormalities allows the identification of functional gene-to-gene relationships involved in similar pathological processes 
. Recent studies conclude that the phenotypic similarity measurement proposed by Robinson and co-workers 
has a significant contribution to the biological coherence compared to text-mining methods 
. Therefore, on the one hand, the study of the similarity among pathologies requires representing them as a set of pathophenotypes instead of a pathological entity. On the other side, pathophenotypic information can be used to reinterpret the relationships among diseases identifying a new pathological phenotypic space that makes it possible the study of novel gene-to-gene associations (as can be seen in the schematic representation in ). Zhang et al. 
have recently stressed some limitations of network-based methods suggesting that the relationships between rare diseases cannot be fully captured by gene-to-gene projections alone. Therefore, the efforts to characterize the genetic and functional environment of given diseases (disease modules) can contribute to enrich the usefulness of disease network analyses.
List of abbreviations used throughout the paper.
In this work, network medicine approaches have been used to study the pathological relationships among genes using semantic similarities (that in this case are pathophenotypic similarities) instead of inferred unipartite edges (gene-to-gene) from bipartite edges (disease-to-gene associations). For instance, a classification of four distinct disease-to-gene associations is proposed () to illustrate possible limitations of the current disease-to-gene network models 
. These classes provide four different subsets of genes in agreement with the number of genes associated with a disease (monogenic or polygenic) and the number of diseases associated with a gene (monotropic and pleiotropic). We have also built a pathophenotypic similarity gene network (PSGN) using semantic similarity 
between genes that are annotated in HPO. The topological features of gene subsets obtained from inferred pathological networks have been analyzed and compared in PSGN. Additionally, the representation of PSGN in three different human biomolecular interactomes based on physical interactions, metabolic flux coupling and functional interactions were also evaluated. For this, a network comparison analysis 
and a subsequent performance validation have been used to study the degree of contribution of each biomolecular interactome to the biological consistency of gene-to-gene pathophenotypic similarities. In addition, this biological coherence can be used to incorporate novel components in disease-causing gene modules, as we demonstrate for maple syrup urine disease (MSUD), an inborn error of the metabolism of branched-chain amino acids.
Summarizing, this work provides evidence that a standard phenotypic profiling expands the genetic disease associations using a specific ontology for human abnormalities. These pathologic relationships among genes were not obvious and, consequently, disregarded in previous disease network analyses.