In pharmaceuticals, strategically identifying new uses for existing drugs—in other words, drug repurposing or indication finding—can accelerate the delivery of therapies to patients and offer substantial cost savings compared with developing novel compounds. However, traditional methods of indication finding have often been slow and limited, relying on expert opinions, literature reviews, and chance observations. Now, AI is stepping in and offering a transformative approach, as explored in the recent paper “Indication Finding: A novel use case for representation learning.”1
In the paper, researchers from McKinsey; Quantum Black, AI by McKinsey; and the Ellison Institute of Technology demonstrate how a specific type of AI, known as “representation learning,” can analyze real-world data to identify potential new uses for existing drugs. This AI technique enables the system to learn complex relationships within vast data sets of patient information. Rather than simply examining individual data points, AI generates “embeddings,” which can be visualized as maps. On these maps, diseases and treatments are positioned near each other based on their similarities and connections. This allows researchers to recognize diseases that might be effectively treated by drugs already used for related conditions.
The researchers focused on anti-IL-17A drugs, which are used to treat inflammatory conditions. They provided the AI with a massive data set of real-world data from more than 17 million patients and asked it to create maps. It then identified diseases that were closely related to the diseases that anti-IL-17A drugs are known to treat.
The AI showed a striking ability to identify diseases such as rheumatoid arthritis, rosacea, and hidradenitis suppurativa for which anti-IL-17A drugs have already shown positive results in clinical trials. In the top 50 indications that the AI ranked, 60 percent were conditions with positive trial results, and none were from conditions for which the drug had failed. In the top 200, all positive-validation conditions were ranked, compared with only 20 percent of those with failed trials. The AI also ranked diseases for which anti-IL-17A drugs have previously failed to demonstrate efficacy, such as Crohn’s disease and atopic dermatitis, much lower on its list. Additionally, it surfaced some diseases for which anti-IL-17A drugs’ efficacy has not yet been tested, pointing to potential new research avenues. This level of accuracy demonstrates that AI can significantly assist in identifying new applications for existing therapies.
While this research is still evolving, the implications for the pharmaceutical industry are profound. By using AI-driven indication finding rather than pursuing lengthy and costly searches for completely new drugs, companies can efficiently identify opportunities to extend the utility of their existing portfolios. This more informed, data-driven decision-making results in faster time to market for new treatments and reduces R&D costs. Furthermore, such strategies can significantly enhance the product life cycle management of existing assets, maximizing their commercial value. Those interested in diving deeper into this research can find the complete paper here.
The original paper was authored by Alex Devereson and David Champagne of McKinsey and Matej Macak of the Ellison Institute of Technology, with Emily Briggs, Ian Lyons, and Jennifer Hou of McKinsey and Alexander Aranovitch, Chris Anagnostopoulos, Maren Eckhoff, and Valmir Selimi of Quantum Black, AI by McKinsey.
1. Maren Eckhoff et al., “Indication finding: A novel use case for representation learning,” QuantumBlack, AI by McKinsey, October 24, 2024.