Preclinical validation of therapeutic targets predicted by tensor factorization on heterogeneous graphs

Abstract

Incorrect drug target identification is a major obstacle in drug discovery. Only 15% of drugs advance from Phase II to approval, with ineffective targets accounting for over 50% of these failures1–3. Advances in data fusion and computational modeling have independently progressed towards addressing this issue. Here, we capitalize on both these approaches with Rosalind, a comprehensive gene prioritization method that combines heterogeneous knowledge graph construction with relational inference via tensor factorization to accurately predict disease-gene links. Rosalind demonstrates an increase in performance of 18%-50% over five comparable state-of-the-art algorithms. On historical data, Rosalind prospectively identifies 1 in 4 therapeutic relationships eventually proven true. Beyond efficacy, Rosalind is able to accurately predict clinical trial successes (75% recall at rank 200) and distinguish likely failures (74% recall at rank 200). Lastly, Rosalind predictions were experimentally tested in a patient-derived in-vitro assay for Rheumatoid arthritis (RA), which yielded 5 promising genes, one of which is unexplored in RA.

Expert-Augmented Computational Drug Repurposing Identified Baricitinib as a Treatment for COVID-19

“…This brought approximately 40,000 additional relationships into the graph. New relationships consisted of disease, biological process, tissue and compound entities, related together through an unsupervised, rule-based model at the sentence level (“SVO,” see Paliwal et al, 2020 ).…”

“…We have described here how we enriched our biomedical knowledge graph ( Paliwal et al, 2020 ) using NLP, using it to identify host biological processes and pathways that are impacted by SARS-CoV-2 infection. Through workflow iterations of graph pattern querying and protein-protein interaction network exploration, a final network was found to be significantly enriched for multiple disease mechanisms of interest, in particular - viral infection and cytokine-mediated inflammation, two processes involved in COVID-19.…”

“…Graph nodes are chiefly biomedical entities, such as diseases, disease processes, pathways, proteins and compounds, with directed relationships scoring various types of connection or association between entities. The KG includes data from both curated databases and literature sources, with its construction described in Paliwal et al (2020) . Importantly, the graph is enhanced for disease processes, mechanisms and pathways, which allow exploration of novel diseases through an understanding of the underlying biology.…”

“…We describe here how, in early 2020, we used a visual analytics approach using interactive computational tools in multiple iterations to identify promising potential treatments. We used BenevolentAI’s drug discovery knowledge graph (KG) of biomedical data ( Paliwal et al, 2020 ) which comprises contents from dozens of biomedical databases, enhanced by information from machine-read scientific literature, covering the enormous amount of biomedical knowledge available (approximately 30 M papers are cataloged in PubMed). Although built for de-novo drug discovery, the inclusion of approved drugs, treatments and other information made it a suitable resource for drug repurposing.…”

Repurposing Vandetanib plus Everolimus for the Treatment ofACVR1-Mutant Diffuse Intrinsic Pontine Glioma

“…The BenevolentAI knowledge graph contains over 40 million documents and over 1 billion relationship edges. It contains diseases, biological tissues, mechanisms and pathways, gene ontology processes, genes and proteins as well as drugs, biologics and small molecules (26).…”