GEXVal's Drug Repurposing Study Using GATE-Based AI Technology Published in Journal “Pharmaceuticals”

株式会社GEXVal
September 25, 2026

GEXVal Publishes COVID-19 Drug Repurposing Study Using Its AI Technology GATE

in the Peer-Reviewed Journal Pharmaceuticals

GEXVal Inc. (President and CEO: Juran Kato, PhD; Location: Fujisawa, Kanagawa, Japan, hereinafter “GEXVal”) is pleased to announce  is pleased to announce that a paper on its AI drug discovery research was published on September 25, 2026, in Pharmaceuticals, a peer-reviewed international journal published by MDPI.

GEXVal was an early mover in applying Graph Attention Autoencoder (GATE)¹⁾, a graph deep learning technology, to drug discovery. GATE is the core technology of RePhaIND^®²⁾, GEXVal’s proprietary next-generation AI drug discovery platform. In this study, GEXVal built a framework that corroborates GATE’s predictions of therapeutic candidates with multi-layered analyses, and validated it using the search for COVID-19 treatments as a case study.

The study was led by GEXVal's Chief Technology Officer (CTO) Yusuke Nakayama, PhD, and conducted in collaboration with Project Associate Professor Shingo Tsuji of the Research Center for Advanced Science and Technology, The University of Tokyo; Professor Kouichi Hosomi of the Faculty of Pharmacy, Kindai University; and others. Dr. Nakayama served as co-first author and corresponding author of the paper.

Drug repurposing starts from compounds for which safety data have already been accumulated. This allows development to proceed quickly while reducing early-stage risk, making it particularly effective when a rapid response is needed, as with emerging infectious diseases. However, predictions from a single computational method are often insufficient to narrow down candidates toward clinical application.

In this study, GATE was trained on a knowledge graph of diseases, drugs, and genes, and identified 16 candidate drugs located close to COVID-19 in the latent space. Twelve of them were already known to be associated with COVID-19 or its symptoms. The remaining four, which had little prior association with COVID-19, were further evaluated through multi-layered analyses: disproportionality analysis of the U.S. FDA Adverse Event Reporting System (FAERS) and pathway analysis using LINCS 2020 gene expression data. As a result, two drugs, cilastatin and megestrol, were selected. For cilastatin, the study also proposes a hypothesized mechanism of action involving DPEP1 inhibition.

We also evaluated GATE’s predictive accuracy by testing whether it could rank known approved drug- –indication pairs among its top candidates. GATE correctly predicted about 80% of these pairs, compared with about 70% for representative existing methods. Based on these results, GEXVal believes GATE is more useful than existing methods for broadly exploring therapeutic candidates.

These findings of this study represent hypotheses based on computational predictions and do not demonstrate efficacy of these drugs against COVID-19. Further experimental validation is required.

[Research Context and Significance]

AI drug discovery broadly falls into two areas: technologies that design new drug candidates, and technologies that identify indications and targets. Our technology belongs to the latter and focuses on identifying new therapeutic opportunities, including drug repurposing. Approaches in this area range from literature mining to and knowledge graph searches. What sets our approach apart is that GATE learns the structure of the knowledge graph itself to explore candidates broadly, and these candidates are then narrowed down by corroborating them with real-world data and gene expression data. This paper presents this framework, which combines exploration and corroboration, in a peer-reviewed publication using COVID-19 as a case study.

Dr. Nakayama, CTO of GEXVal, stated: "In this study, we presented a framework that evaluates AI predictions through multiple layers of evidence from real-world data and phenome-level analysis. We believe this approach of building up evidence to narrowing down candidates that a single analysis might miss can be applied not only to COVID-19 but also to rare diseases and other conditions with few treatment options. Based on these results, we will further advance exploration with RePhaIND^®, our AI drug discovery platform."

**Publication Details**

[About Pharmaceuticals]
Pharmaceuticals is an international, peer-reviewed, open access journal in medicinal chemistry and pharmacology published by MDPI (Basel, Switzerland) since 2004. It is indexed in PubMed, Scopus, and Web of Science (SCIE). This paper was published in the journal's "AI in Drug Development" section.

URL: https://www.mdpi.com/journal/pharmaceuticals

[About GEXVal]

GEXVal strives to create and develop innovative pharmaceuticals for unmet medical needs, ensuring Treatment Reaching the Unreached with focus on rare diseases and underserved medical conditions. By leveraging our proprietary AI-powered pharmacoinformatics technology, we illuminate paths to breakthrough therapies, identifying hidden potential in drug candidates to deliver life-changing medicines that bring new hope to patients and their families.

1) About Graph Attention Autoencoder（GATE）
Graph Neural Networks (GNNs) are AI technologies that learn patterns from various "connections" in data. In particular, Graph Attention Network (GAT) is widely used in everyday services such as social media recommendations and logistics route optimization. Notably, Graph Attention Autoencoder (GATE), which GEXVal pioneered in drug discovery, has previously been limited to specific applications in life sciences, such as protein structure prediction. By innovatively applying GATE technology to the entire drug discovery process, GEXVal enables the discovery of novel therapeutic candidates that would otherwise remain unidentified through conventional methods.

2) About RePhaIND^®：Revolutionary Pharmacoinformatics to Find IND (Investigational New Drug)
RePhaIND^® is GEXVal's proprietary AI-driven drug discovery platform with three key features:
EMPOWER: Discover hidden relationships between drug candidates and diseases
ACCELERATE: Dramatically improve efficiency compared to conventional methods
REVOLUTIONIZE: Enable new approaches to unmet medical needs, particularly in rare diseases, bringing a paradigm shift to the drug discovery process
RePhaIND^® is a registered trademark of GEXVal Inc. in Japan, China, Hong Kong, Korea, Australia, Europe, UK, and US and a trademark in other countries and regions.

有关本事宜请咨询:
Head of Corporate Office
Atsushi Sugizaki
info@gexval.com