Harnessing AI and C. elegans: Bridging Prediction and Proof in Drug Discovery
The past decade has seen an explosion of companies employing artificial intelligence (AI) for drug discovery and development (AIDD). Although different corporations possess their own proprietary algorithms, the basic principle remains the same: machine-learning algorithms are trained on existing biomedical data to identify patterns, correlations, and trends hiding in plain sight. Traditional methods of drug discovery rely on trial-and-error medicinal chemistry and large-scale screening. Promising candidates might bear objective similarity to one another, but these relationships might not be immediately obvious to a human observer.
AIDD companies typically adopt one of three models, differing in where the ‘unknowns’ are found in the pipeline1.
- Known drugs, unknown targets: AI is used to generate new hypotheses associating diseases with specific targets – whether these are genes, proteins, or other molecules. Known drugs or reagents are then repurposed to try and modulate the chosen target. This approach has promising implications for rare diseases, 95% of which have no known treatments2. As these drug candidates already possess an existing safety profile, this approach can accelerate clinical trial progress to phase II. This method’s main drawback is the reduced likelihood of successfully striking a target.
- Known drugs, known targets: AI-driven design is used to improve the efficacy of established drugs acting on established targets. This removes the obstacle of target identification and validation but introduces the risk of being outcompeted by larger companies with deeper pockets.
- Unknown drugs, unknown targets: This approach features end-to-end AIDD, where machine learning is integrated into every aspect of the development lifecycle, ensuring novelty for both target and compound. This tactic is high risk, both in effective target identification and in de novo drug design, but has the potential to establish a candidate without any competitors on the market.
This list is by no means exhaustive. In theory, a machine learning algorithm can be tacked on to every phase of drug development, including identification of biomarkers, literature revision, and post-market surveillance3,4. AI-driven multiomics analyses of participants before a clinical trial can also help researchers assess whether a subject is likely to respond to a drug, and why.
Most AIDD companies claim greater speed, breadth, and success rates than traditional drug discovery methods - but how true is this? As it stands, no AI-developed drugs have progressed beyond phase II trials4.
The AI + C. elegans drug discovery pipeline
One of AIDD’s greatest pitfalls is the paucity of high-quality training data. Even if a company can circumvent acquisition costs and privacy regulations, a fundamental lack of published ‘negative’ clinical data can occlude drug-target-disease associations. Another major limitation of AI-driven drug discovery is its tendency to produce false positives - compounds that look promising computationally but fail in biological testing. Many AI models operate as 'black boxes,' making it difficult to pinpoint why a particular compound was selected. Without mechanistic insight, researchers may struggle to refine their predictions and validate AI-generated candidates effectively.
Although AI can help to generate better-informed ideas in the form of computational predictions, rigorous laboratory validation is still required at every checkpoint, and developers may find themselves in the same regulatory bottlenecks that they sought to avoid.
One way of mitigating these limitations is to validate therapeutic candidates in preclinical models where disease states can be genetically simulated, like the fruit fly Drosophila melanogaster5 or the nematode Caenorhabditis elegans6. This approach has been applied not only to disease but to generalised physical conditions such as ageing7. A recent study published in Nature Aging used an AI prediction platform to identify novel geroprotectors (compounds that promote healthy ageing). When researchers at the Indian Institute of Technology identified geroprotective candidates outside of their training data, the group validated their capacity to extend lifespan in yeast and in C. elegans. AI excels at predicting geroprotective compounds, but prediction alone is not enough - these compounds must demonstrate real-world efficacy in biological models. The researchers tested their AI-selected compounds in C. elegans, a powerful model for lifespan and healthspan studies, before moving on to senescence and toxicity assays in human fibroblasts. This multi-model validation approach strengthens confidence in the translatability of these compounds. This validation step is crucial for weeding out false positives and prioritizing the most promising candidates for further testing. These models offer the benefit of rapid data generation as well as well-conserved ageing-associated pathways, allowing them to be used in tandem and the researchers found that their model was able to successfully predict endogenous metabolites that delay senescence in human fibroblasts6.
Magnitude Biosciences provides in vivo validation after AI-informed drug discovery, and our C. elegans powered WormGazer™ technology offers a rapid, scalable way to test AI-generated predictions, reducing false positives before they reach costly mammalian trials. Compared to other preclinical models, C. elegans offers unique advantages - conserved ageing pathways, rapid life cycle, and high-throughput screening capabilities - making it an ideal first-pass system for AI-discovered compounds. See our technology page to find out more.
References:
1 Wilczok, Dominika, and Alex Zhavoronkov. 2024. “Progress, Pitfalls, and Impact of AI Driven Clinical Trials.” Clinicial Pharmacology and Therapeutics 1-4.
2 Korsunska, A, SE Bolden, M Repasky, M Zuccato, and D Fajgenbaum. 2023. “The ROADMAP Project.”
3 Blanco-González, A, A Cabezón, A Seco-González, D Conde-Torres, P Antelo-Riveiro, A Piñeiro, and R Garcia-Fandino. 2023. “The Role of AI in Drug Discovery: Challenges, Opportunities, and Strategies.” (Pharmaceuticals (Basel)).
4 Zhang, K, X Yang, Y Wang, Y Yu, N Huang, G Li, X Li, JC Wu, and S Yang. 2025. “Artificial intelligence in drug development.” (Nature Medicine).
5 Taylor, P. 2023, July 13. Consortium hunts new KRAS drugs with fruit flies and AI. https://pharmaphorum.com/news/consortium-hunts-new-kras-drugs-fruit-flies-and-ai
6 Arora, S, A Mittal, S Duari, S Chauhan, NK Dixit, SK Mohanty, A Sharma, et al. 2024. “Discovering geroprotectors through the explainable artificial intelligence-based platform AgeXtend .” (Nature Aging)