Building a knowledge graph to enable precision medicine

Abstract

Developing personalized diagnostic strategies and targeted treatments requires a deep understanding of disease biology and the ability to dissect the relationship between molecular and genetic factors and their phenotypic consequences. However, such knowledge is fragmented across publications, non-standardized repositories, and evolving ontologies describing various scales of biological organization between genotypes and clinical phenotypes. Here, we present PrimeKG, a multimodal knowledge graph for precision medicine analyses. PrimeKG integrates 20 high-quality resources to describe 17,080 diseases with 4,050,249 relationships representing ten major biological scales, including disease-associated protein perturbations, biological processes and pathways, anatomical and phenotypic scales, and the entire range of approved drugs with their therapeutic action, considerably expanding previous efforts in disease-rooted knowledge graphs. PrimeKG contains an abundance of ‘indications’, ‘contradictions’, and ‘off-label use’ drug-disease edges that lack in other knowledge graphs and can support AI analyses of how drugs affect disease-associated networks. We supplement PrimeKG’s graph structure with language descriptions of clinical guidelines to enable multimodal analyses and provide instructions for continual updates of PrimeKG as new data become available.

Multimodal representation learning for predicting molecule–disease relations

“…The embedding vectors are first reduced to two dimensions via PCA and then visualized using t-SNE ( Van der Maaten and Hinton, 2008 ). We visualize 46 cancer drugs ( https://www.cancer.gov/about-cancer/treatment/drugs ) and 26 psychotropic drugs ( https://www.healthpartners.com/ucm/groups/public/@hp/@public/documents/documents/entry_194823.pdf ), which are commonly used in clinical settings, and their related top-30 indications reported on PrimeKG dataset ( Chandak et al , 2022 ) in Figure 2a and their top-30 reported side effects on the SIDER 4.1 dataset ( Kuhn et al , 2016 ) in Figure 2b . Meanwhile, we also visualize all EHR concepts and Drugbank molecules in Supplementary Figure S3a , showing that they are well-aligned in the embedding space.…”

“…We experiment on PrimeKG ( Chandak et al , 2022 ), which includes annotated drug-indication relations and SIDER 4.1 ( Kuhn et al , 2016 ), which consists of reported side effects for joint training. There are 349 indications included that are observed by at least three drugs in PrimeKG.…”

Drug Synergistic Combinations Predictions via Large-Scale Pre-Training and Graph Structure Learning

“…-PrimeKG [36]: A knowledge graph dataset integrating 20 high-quality datasets, biorepositories, and ontologies.…”

“…Feature Representation Unsupervised learning shows brilliant results in the field of Natural Language Processing and bioinformatics, many sequence-based biological tasks benefit from it. Inspired by the progress of unsupervised frameworks, we take advantage of three models with large-scale information: ESM-1b [34], KPGT [33], and PrimeKG [36] to obtain the initial embedding of our protein, drug, and disease respectively. Fig.…”

“…Meanwhile, the model generates embeddings for every protein in the expression by utilizing a protein language model [34]. To enrich more features, we also include disease information, particularly, we apply RotatE [35] to get the embedding of disease from PrimeKG [36]. Next, we utilize graph neural networks and take our generated embeddings as node representations.…”