Human phenotype ontology: developing hpo terms for mood disorders to integrate diverse phenotypic datasets

Psychiatric genetics has been limited by a lack of descriptive phenotypic standards. Diagnostic standards such as the ICD and DSM are widely used, but only provide broad labels for heterogeneous behavioral phenotypes. Such diagnostic labels are unable to describe the intrinsically fluid phenotypic landscape of behavioral disorders, which present along multidimensional behavioral spectra and are influenced by a multitude of genetic and environmental factors. A phenotypic standard for granular behavioral features could describe this phenotypic landscape and facilitate detailed analysis of multisite datasets such as those provided by the PGC. To create such a standard, we are expanding the Human Phenotype Ontology (HPO) – a widely accepted standard for describing symptoms of somatic disorders – to define and relate granular observable phenotypic features of behavioral disorders. Created in 2008, the HPO now includes over 16,000 terms organized in an ontological hierarchy such that the HPO branches from a parental (root) term to increasingly specific child terms. The HPO has been adopted by consortia such as the 100,000 Genomes Project and SOLVE-RD, demonstrating the practicality of the HPO's structure for genetic analysis. Unlike existing behavioral phenotyping standards, the HPO identifies atomic phenotypic observables and connects them according to their meaning rather than their patterns of co-occurrence (as in DSM). The HPO's hierarchical structure allows the comparison of datasets with varying phenotypic resolutions. For example, one clinical site may evaluate depression with the MADRS, another with the PHQ-9, and still another with the QIDS. While the MADRS asks about insomnia, the QIDS asks more specifically about terminal, maintenance, and onset insomnia. The PHQ-9 is more general asking only about sleep disturbance. When comparing these data, we are confronted with misaligned phenotypic features. The HPO addresses this by encoding IS-A relationships among features. Feature A IS-A Feature B if A is a subclass of B, i.e., any manifestation of A is also a manifestation of B. Terminal insomnia IS-A Insomnia, Insomnia IS-A Sleep disturbance. An analysis using HPO terms can infer when the QIDS records (for example) terminal insomnia that the patient also has the more general phenotypes of insomnia and sleep disturbance which appear on the MADRS and PHQ-9. Such semantic encodings allow for seamless integration of datasets with varying granularity without needing to coerce phenotypic observations into categorical constructs. In 2022, the HPO had 215 behavioral terms. To make the HPO more useful for behavioral phenotyping we have begun creating HPO terms to describe features of mood and neurodevelopmental disorder with plans to extend this work to all behavioral disorders. This process has involved working with cohorts of domain experts who identify and define important terms to be added to the HPO. New terms are placed in the HPO hierarchy according to their IS-A relationships with existing terms. Each term is reviewed by several experts to ensure that the definition is robust for diverse users. These efforts have resulted in the creation of a hundred terms that describe behaviors common in autism spectrum disorder through an international workshop at the National Institute of Mental Health. We are continuing to enhance the coverage of behavioral disorders by developing HPO terms for mood disorders in conjunction with the Global Bipolar Consortium.