Integrating and formatting biomedical data as pre-calculated knowledge graph embeddings in the Bioteque

Abstract

Biomedical data is accumulating at a fast pace and integrating it into a unified framework is a major challenge, so that multiple views of a given biological event can be considered simultaneously. Here we present the Bioteque, a resource of unprecedented size and scope that contains pre-calculated biomedical descriptors derived from a gigantic knowledge graph, displaying more than 450 thousand biological entities and 30 million relationships between them. The Bioteque integrates, harmonizes, and formats data collected from over 150 data sources, including 12 biological entities (e.g., genes, diseases, drugs) linked by 67 types of associations (e.g., ‘drug treats disease’, ‘gene interacts with gene’). We show how Bioteque descriptors facilitate the assessment of high-throughput protein-protein interactome data, the prediction of drug response and new repurposing opportunities, and demonstrate that they can be used off-the-shelf in downstream machine learning tasks without loss of performance with respect to using original data. The Bioteque thus offers a thoroughly processed, tractable, and highly optimized assembly of the biomedical knowledge available in the public domain.

Prioritization of new candidate genes for rare genetic diseases by a disease-aware evaluation of heterogeneous molecular networks

“…Thus, RD diagnosis can become a “needle in a haystack” task where the solution is still far connected to current knowledge. In consequence, to augment their efficiency in the difficult cases, the gene-disease prediction methods are encouraged to use new and various types of functional annotations and combine them accurately (15, 16, 61, 62). We first confirmed that different types of gene-gene functional association networks had different capabilities in recovering known genes associated to a large collection of RDs.…”

“…If unavailable, edge weights are all set up to 1 (Table S1). Thus, for networks describing human gene phenotype similarity using HPO (32) and phenotype similarity using mouse orthologs from the Mouse Genome Informatics (MGI) (16) we calculated Jaccard similarity for each pair of genes sharing at least one HPO term and constructed a null distribution of Jaccard values to compute z-scores. Significant gene interactions (z-score>1.96; p-values<0.05) were selected to generate the network of phenotypes.…”

“…Alternative strategies were run across 63 PanelApp GAG gene sets and validated in RG genes. Recall-at-k was measured to quantify what fraction of all the disease genes are ranked within the first k predicted genes (k = 8, 16,32,64,128,256,512). Best integration methods are selected at recall-at-n (validation set size).…”

“…They have in common the ability to screen large datasets to extract hidden associations. These sources can be of different nature, from literature to the different flavors of omics data, including their integration (5, 7) which has been reported to augment accuracy (15, 16). On the top of this, resources like PanelApp (17) provide manually curated gene candidates for genetics diseases based on the community feedback.…”

Advancing Targeted Protein Degradation via Multiomics Profiling and Artificial Intelligence

“…The challenges are exacerbated when input data are collected from multiple sources in the public domain. To ease this task, large biological knowledge graphs have been suggested as an intuitive framework for data integration, − followed by graph embedding techniques capable of expressing the context of each entity (e.g., a compound or a gene) in a dense vectorial format. In brief, vector “embeddings” capture the interactions and statistical correlations found in the knowledge graph, such that highly connected nodes will have similar vector embeddings.…”