Dr Anton Yuryev, consulting director, Bioinformatics and Data Science, Elsevier, explores two case studies where researchers have used AI to find drugs that can be repurposed for rare conditions.
There are more than 300 million people worldwide living with a rare condition, but only 5% of 8,000 known rare diseases have licensed treatments. One pragmatic and potentially life-changing solution to the rare disease challenge is drug repurposing – the exploration of new uses for approved, discontinued and investigational drugs.
Data from Elsevier’s Scopus finds this approach has risen in popularity in recent years; in 2010, only one published document covered drug repurposing for rare diseases, by 2024 that had risen to 72.
Now, advanced AI tools are making repurposing more accessible than ever. AI is allowing researchers to rapidly and accurately analyse large, disparate datasets and discover hitherto unknown connections – tasks that, in the past, might have taken humans many months to achieve by themselves.
Two case studies showcase the power of this approach:
1. A collaborative project between Elsevier and the Sinergia Consortium brought together data and disease experts to create in silico molecular pathway models for diffuse intrinsic pontine glioma (DIPG), a rare form of childhood brain cancer. The complex and poorly understood maturation pathways of human neuron precursors had limited treatment discovery for DIPG. Researchers successfully used AI techniques to better understand these pathways and identify drug repurposing candidates. The project has also developed recommendations for selecting drugs based on the type of mutation in TP53 gene; this gene is mutated in 50% of all cancers.
2. A SciBite and Stardog partnership used AI to uncover drug repurposing candidates to treat Friedreich’s ataxia, a rare genetic disorder caused by a deficiency of frataxin – a small nuclear-encoded mitochondrial protein. By using AI-powered semantic enrichment on the data, the tools were able to find candidates for repurposing in just 40 minutes.
These projects not only demonstrate the potential of AI in rare disease research, but also the best practices for its successful implementation.
AI-powered text mining
One area where AI excels is its ability to mine huge volumes of data and reveal associations. This is particularly useful in rare diseases, where research is often scattered across many different sources and documents – such as journals, EHRs, trial results and conference papers. AI can mine and extract insights and turn the text results into structured data.
The researchers in the DIPG example began by analysing ‘omics data from real-world patients to identify the most active genes in DIPG. They focused on cancer suppressor TP53, one of the most frequently mutated genes in cancer cells that drives radio resistance. They used AI to text-mine PubMed abstracts and full-text journal articles for reports of drugs inhibiting “mutant TP53”.
From there, the researchers compiled a list of 144 approved drugs known to target this mutation, scoring candidates according to Gene Set Enrichment Analysis (GSEA) enrichment score and by the number of inhibited regulators.
By analysing the mechanisms for known mutant TP53 inhibitors, the AI was able to predict an additional 322 drugs that might be able inhibit mutant TP53. A combination of human expertise and AI analysis identified five drugs as being the most promising for further research.
The value of FAIR data sources
AI cannot work optimally without access to multiple sources and formats of trusted data. Examples of potential data sources include disease pathology information (e.g., pharmacokinetic, efficacy, and metabolising enzyme and transporter data), drug compound libraries and databases, clinical trials data, ‘omics data (e.g., genomics, proteomics, phenomics insights) and regulatory and approval documentation (e.g., FDA/EMA documents).
Crucially, all data must be FAIR – Findable, Accessible, Interoperable, and Reusable – to ensure any AI model is working from the strongest foundation. The Friedreich’s ataxia research team ensured their data was FAIR by embedding AI ontology creation into the process.
Ontologies are expert-informed frameworks that define and organize concepts, entities, and relationships in a specific domain – enabling standardized terminology, data integration, and semantic search. Ontologies create a shared vocabulary that allows researchers to make sense of unstructured text.
Friedreich’s ataxia is a neurodegenerative disease caused by reduced expression of the mitochondrial protein frataxin, and the researchers were in search of drugs known to increase frataxin impression.
Using an AI-powered entity recognition engine to expand the synonyms from ontologies, the team searched the text for synonyms and identified additional entities of interest.
All entities were mapped back to a common identifier term. The team then rapidly extracted data from 4,000 documents in under an hour to create a knowledge graph visualising how certain drugs increase the expression of frataxin in Friedreich’s ataxia.
The knowledge graph connected data in a way that was not possible manually or in siloed systems by combining both data and ontology-driven metadata, enabling a simple way for researchers and AI tools to query the data. The researchers were able to pinpoint four Phase IV drugs that increased the expression of frataxin.
Gathering a range of voices and expertise
As the two examples demonstrate, drug repurposing projects require a combination of not just AI technology, but also data science and disease and domain experts. Domain experts provide the context necessary to parse research questions for AI, inform ontologies, and understand which datasets are needed. Skilled data scientists can standardise, harmonise and semantically enrich data to ensure it is machine-readable.
It’s also important to remember that patients are a valuable source of data and insight, providing additional context into a condition when literature is limited. Rare disease patient groups can shed light on disease symptoms and progression and offer access to the right patient cohort for trials.
A new era of rare disease research
It’s clear that combining AI with human expertise has huge benefits in accelerating and streamlining complex scientific workflows in rare disease research, with AI’s ability to process and analyse ‘needle in the haystack’ data. These case studies demonstrate that AI-enhanced drug repurposing is one of the most promising developments in rare disease research – with the potential to greatly aid patients in need of new treatments.
