Computational Frameworks for Drug Repurposing

108 concepts95 papers1 proposal

The traditional pipeline for developing a new drug from scratch is notoriously inefficient. It requires over a decade of research, billions of dollars in investment, and faces an incredibly high failure rate in late-stage clinical trials. To bypass these barriers, researchers are increasingly turning to drug repurposing (or repositioning)—the strategy of identifying new therapeutic indications for existing, approved medications.

Because these compounds have already passed rigorous safety and toxicity screenings, repurposing drastically compresses the development timeline and reduces regulatory hurdles. However, finding a new use for an old drug historically relied on serendipity. Today, modern drug repurposing has evolved into a systematic, data-driven discipline. By leveraging machine learning, network biology, and transcriptomics, computational frameworks can proactively map the hidden relationships between drugs, biological targets, and diseases.


The Network Perspective: Proximity in Biological Graphs

To understand why a drug approved for one disease might treat another, we must move away from the "one drug, one target, one disease" paradigm. Biological systems are highly interconnected networks. Diseases are rarely caused by a single isolated gene; instead, they manifest as disruptions within a localized neighborhood of a protein-protein interaction (PPI) network, known as a disease module.

Computational frameworks exploit this topology using the principle of "guilt by association." If a drug targets a protein that sits in close proximity to, or shares pathways with, a specific disease module, it can exert a therapeutic effect on that disease—even if the drug was designed for an entirely different organ system.

[Drug] ---> (Primary Target) --shared pathway--> (Disease Module Protein) ---> [New Indication]

For example, Haoyang Wu et al. developed a network-based framework called MD-Miner (MD-Miner: a network-based approach for personalized drug repositioning) that maps a drug’s molecular mechanism directly onto a patient's specific disease signaling network. Rather than looking for simple gene matches, this approach traces how a drug's influence propagates through the network, achieving a significantly higher success rate in identifying effective therapies compared to random screening.

To scale this up, researchers model these interactions as massive heterogeneous graphs containing multiple node types (drugs, diseases, genes, pathways) and edge types (binds, upregulates, associates). Advanced machine learning models then predict new indications by framing the problem as a link prediction task.

For instance, Yuxing Wang et al. proposed GeneDR (Gene based message passing for drug repurposing), which uses Long Short-Term Memory (LSTM) networks to pass messages along specific "gene paths" before feeding the information into a Graph Convolutional Network (GCN). By learning the sequential biological context of how signals flow from a drug to a disease, the model accurately predicts missing drug-disease links. Similarly, Maryam Lotfi Shahreza et al. utilized a label propagation algorithm on a heterogeneous network (A computational drug repositioning method applied to rare diseases: Adrenocortical carcinoma) to uncover novel treatments for rare orphan cancers, demonstrating how network topology can overcome the lack of extensive clinical data.


The Transcriptomic Perspective: Signature Inversion

While network biology looks at the structural blueprint of cells, transcriptomics captures their functional state. When a tissue is affected by a disease, its gene expression profile changes—some genes are abnormally upregulated, while others are downregulated. This shift constitutes a transcriptomic "fingerprint" of the disease.

The core intuition behind transcriptomic drug repurposing is signature inversion: if a disease pushes the cellular state in direction $X$, we want to find a drug that pushes it in the exact opposite direction $-X$. A drug whose transcriptomic signature is highly anti-correlated with the disease signature acts as a molecular counterweight, restoring the cell to a healthy equilibrium.

Disease Signature:   [↑ Gene A]  [↓ Gene B]  [↑ Gene C]
                        |           |           |
                     (Opposite / Anti-correlated)
                        |           |           |
Drug Signature:      [↓ Gene A]  [↑ Gene B]  [↓ Gene C]  ===> Potential Therapeutic!

This approach is highly effective for tackling complex or drug-resistant conditions. Kelly Regan et al. leveraged this mechanism to address sorafenib-resistant hepatocellular carcinoma (Transcriptomics-Based Drug Repurposing Approach Identifies Novel Drugs against Sorafenib-Resistant Hepatocellular Carcinoma). By analyzing the transcriptomic profiles of resistant cancer cells, the authors computationally screened drug libraries to find compounds capable of reversing the resistance signature, successfully validating two candidates in vitro.

To refine this process, researchers often restrict their analysis to tissue-specific pathways to minimize off-target noise. Liang Yu et al. designed a pipeline that focuses exclusively on liver-specific metabolic and signaling pathways (Predicting therapeutic drugs for hepatocellular carcinoma based on tissue-specific pathways), identifying securinine as a potent anti-cancer agent with high selectivity for tumor cells over healthy tissue. By combining these transcriptomic fingerprints with deep learning and chemical structure analysis, as demonstrated by Bingrui Li et al. (A novel drug repurposing approach for non-small cell lung cancer using deep learning), models can capture both the structural capability of a drug and its downstream biological impact.


The Similarity Perspective: Exploiting Polypharmacology

Many drugs are "dirty"—they bind to multiple biological targets beyond the primary one they were originally engineered to hit. This phenomenon, known as polypharmacology, is a major driver of drug repurposing. If two drugs share a significant number of secondary targets or exhibit high structural similarity, they are likely to share therapeutic effects.

By constructing Drug-Drug Similarity Networks, researchers can group small molecules into functional communities based on their interaction profiles rather than their intended clinical use. When a drug clusters tightly within a community of medications dedicated to a completely different therapeutic area, it signals a hidden capability.

Lucreția Udrescu et al. demonstrated this by clustering drug-drug interaction networks using energy-model layouts (Clustering drug-drug interaction networks with energy model layouts: community analysis and drug repurposing). The authors found that while 85% of drugs clustered according to their known therapeutic classes, the remaining 15% appeared "out of place." These topological anomalies represent prime candidates for repurposing. In subsequent work, the authors shifted from chemical structures to target-based similarity networks (Uncovering new drug properties in target-based drug-drug similarity networks), successfully identifying non-cancer drugs that possess potent anti-tumor properties due to shared underlying target profiles.

Complementing this, Guohua Huang et al. combined chemical-chemical interactions with structural similarities (Prediction of Drug Indications Based on Chemical Interactions and Chemical Similarities) to build a multi-feature predictor. This hybrid approach ensures that even if a drug's primary mechanism is unique, its structural fragments or interaction patterns can reveal its utility for alternative indications.


Connecting Silos via Knowledge Graphs and Literature Mining

A significant barrier in biomedical research is the isolation of knowledge. Discoveries regarding a drug's metabolic side effects might be published in a pharmacology journal, while a novel disease pathway might appear in a genetics journal. Computational frameworks break down these silos by aggregating unstructured text and structured databases into unified Knowledge Graphs (KGs).

Using Literature-Based Discovery (LBD) and natural language processing, algorithms extract semantic triples (e.g., Drug X $\rightarrow$ inhibits $\rightarrow$ Protein Y) from millions of scientific abstracts. Once a knowledge graph is constructed, neural knowledge graph completion models can infer missing relationships that no single human researcher could deduce from the literature alone.

[Paper A] ---> (Drug X inhibits Protein Y) \
                                            ===> [Knowledge Graph] ===> Inferred Link: Drug X treats Disease Z
[Paper B] ---> (Protein Y drives Disease Z) /

During the pandemic, Dimitar Hristovski et al. combined literature mining with neural knowledge graph completion (Drug repurposing for COVID-19 via knowledge graph completion) to rapidly surface existing drugs capable of modulating the inflammatory cytokine storm. Similarly, Yi Nian et al. trained graph embedding models on millions of biomedical triples (Mining on Alzheimer's diseases related knowledge graph to identity potential AD-related semantic triples for drug repurposing) to uncover novel therapeutic candidates for Alzheimer's disease, demonstrating how automated reasoning over existing literature can reveal unexpected connections between disparate fields.

By unifying network topology, transcriptomic signatures, and semantic knowledge, computational drug repurposing transforms medicine from a process of slow, deliberate discovery into an agile, predictive science.

Key Foundational Work

Mouse model phenotypes provide information about human drug targets (2014) — contributed 2 concept elements to this direction.
Network-Based Inference Methods for Drug Repositioning (2015) — contributed 1 concept elements to this direction.
Prediction of Drug Indications Based on Chemical Interactions and Chemical Similarities (2015) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_drug_repurposing_6b8bb92e,
  title = {drug repurposing},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/drug-repurposing/rdr_6b8bb92e8019449e8155a04cbe303f06},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-21},
}

This exact version (version 1, frozen text):

@misc{lacuna_drug_repurposing_6b8bb92e_v1,
  title = {drug repurposing},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/drug-repurposing/rdr_6b8bb92e8019449e8155a04cbe303f06/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-21},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake