Computational analysis of neurodevelopmental phenotypes: Harmonization empowers clinical discovery

Abstract

Making a specific diagnosis in neurodevelopmental disorders is traditionally based on recognizing clinical features of a distinct syndrome, which guides testing of its possible genetic etiologies. Scalable frameworks for genomic diagnostics, however, have struggled to integrate meaningful measurements of clinical phenotypic features. While standardization has enabled generation and interpretation of genomic data for clinical diagnostics at unprecedented scale, making the equivalent breakthrough for clinical data has proven challenging. However, increasingly clinical features are being recorded using controlled dictionaries with machine readable formats such as the Human Phenotype Ontology (HPO), which greatly facilitates their use in the diagnostic space. Improving the tractability of large-scale clinical information will present new opportunities to inform genomic research and diagnostics from a clinical perspective. Here, we describe novel approaches for computational phenotyping to harmonize clinical features, improve data translation through revising domain-specific dictionaries, quantify phenotypic features, and determine clinical relatedness. We demonstrate how these concepts can be applied to longitudinal phenotypic information, which represents a critical element of developmental disorders and pediatric conditions. Finally, we expand our discussion to clinical data derived from electronic medical records, a largely untapped resource of deep clinical information with distinct strengths and weaknesses.

Learning with phenotypic similarity improves the prediction of functional effects of missense variants in voltage-gated sodium channels

“…Beyond the gain in predictive performance, this method has several distinct advantages: i) HPO terms have become the standardized language in clinical genetics. They have been used to structure phenotypic information in large-scale sequencing consortia (42), enable deep phenotyping at the point of care (48), and drive phenotype matching software such as Exomiser (32); ii) Conventionally, all features used for model training must be available at time of prediction. That is, presence or absence of all clinical information would have to be explicitly coded which quickly becomes intractable for complex models.…”

“…Additionally, SCN9A was learnt with low phenotypic weight likely due to class imbalance as the training data only included a single SCN9A -LOF. Phenotypic learning will benefit from large deeply phenotyped cohorts and more granular representation of concepts within the ontology (42).…”

Enriching Representation Learning Using 53 Million Patient Notes through Human Phenotype Ontology Embedding

“…We have contributed to HPO terminology since 2010 [4][5][6][7]. The HPO has been widely used for harmonization of clinical features in various studies, including, but not limited to, semantic unification of common and rare diseases [8], genetic discoveries in pediatric epilepsy [9,10], and delineation of longitudinal phenotypes [11,12]. In addition, the HPO is commonly used for genomic studies and allows for analyses of clinical data at a scale that is required by current and future initiatives.…”

“…Comparing phenotypes to one another is a common building block for downstream clinical tasks. Methods for measuring phenotypic similarities, using measures such as the Resnik score [22], information coefficient [23], and graph information content [24], show promise for diagnoses and open doors to novel biological insights such as genetic discoveries [9, 25, 26]. However, these methods are not generally transferable to other tasks and require a significant amount of computation, even with minor changes to the data.…”

“…However, these methods are not generally transferable to other tasks and require a significant amount of computation, even with minor changes to the data. As a result, while these methods perform relatively well on limited data with hundreds of patients, they are computationally intensive for large-scale data with millions or even thousands of patients with diverse clinical representations [18, 26].…”