Artificial intelligence and machine learning are rapidly reshaping how pharmaceutical companies discover and prioritise new compounds [1]. From virtual screening and generative chemistry to predictive toxicology and target identification, AI-driven workflows are now embedded across early-stage R&D. But AI is only as good as the real biological data it is trained on. As these approaches mature, a key question has emerged: how can AI-first discovery pipelines be grounded in robust, biologically meaningful evidence early enough to reduce downstream risk? This is where whole-organism high-throughput screening (HTS) offers unique and under-utilised value.  

VivoScan™, Magnitude Biosciences’ C. elegans-based lifespan and healthspan HTS platform, is designed to integrate seamlessly into AI-enabled drug discovery workflows, providing scalable, early, organism-level data that complements, strengthens, and de-risks machine learning driven decisions. C. elegans is a millimetre-long nematode worm widely used as a model organism, with a fully mapped biology and a decades-long track record in genetics, ageing, and drug discovery [2]. With this tiny organism, Magnitude Biosciences aims to turn in-vivo healthspan and lifespan data into an abundant commodity for the first time ever. 

The rapid lifespan, low cost, ease of handling, and minimal ethical burden of C. elegans make it the only well-established model organism suited to true high-throughput in vivo screening - especially for lifespan and healthspan effects. No other organism combines sufficient biological complexity with the speed, scale, and practicality required for early-stage whole-organism testing. As a result, VivoScan™ represents not just a practical solution for whole-organism screening, but the only viable way to generate organism-level training data at scale.  

AI excels at pattern recognition, optimisation, and prioritisation. Given sufficient data, models can rank compound libraries, explore chemical space efficiently, elucidate structure–activity relationships, and predict molecular properties at scale. However, AI systems are fundamentally limited by the biological scope of their training data. Most models are trained on in vitro assays, biochemical endpoints, omics datasets, or historical clinical outcomes. While powerful, these data sources often lack integrated organismal responses, which limits their ability to predict longitudinal outcomes or systemic toxicity and pleiotropy. As a result, AI pipelines can confidently prioritise compounds that later fail due to effects that only emerge at the whole-organism level [3]. For example, a generated protein may fold correctly according to AlphaFold but may not express well in cells, could be insoluble, or could be immunogenic. 

VivoScan™ addresses this gap by generating high-throughput, quantitative, whole-organism data early in the discovery process. Using C. elegans as a validated model organism [4], VivoScan™ captures lifespan effects, healthspan trajectories, dose-dependent toxicity, and trade-offs between efficacy and organismal fitness. Crucially, this data is scalable, standardised, longitudinal, and machine-readable. Far from competing with AI/ML approaches, VivoScan™ functions as an upstream biological intelligence layer, feeding organism-level reality into model-driven pipelines.  

VivoScan™ integrates seamlessly into the closed-loop optimisation pipelines that modern AI-enabled drug discovery increasingly relies on. When generative models propose novel small molecules, repurposing candidates, structure modifications or drug targets, VivoScan™ enables rapid high-throughput evaluation for lifespan, healthspan, and toxicity effects in a whole organism. Hits are separated from false positives, and results are fed back into ML models to retrain predictors. This loop supports both active learning and multi-objective decision-making (e.g. efficacy vs toxicity vs longevity impact). The result is less time wasted on spurious hits, and faster convergence towards biologically accurate models.  

One of the central challenges facing AI-led drug discovery is late biological failure. Compounds optimised in silico or in vitro often encounter unexpected toxicity, particularly in whole organisms where complex multi-system interactions may occur. By introducing whole-organism screening at the earliest feasible point in the discovery pipeline, VivoScan™ enables false-positive candidates to be filtered out sooner, while providing AI models with rich, longitudinal in-vivo data that cannot be generated at comparable scale using any other model organism.  

This aligns with a core industry goal: fail earlier, fail cheaper, and learn more from every failure.  

Importantly, VivoScan™ is model-agnostic. It does not require specific AI architectures, proprietary algorithms, or changes to internal data science teams. Model architectures will evolve, training pipelines will change, but the need for high-quality organismal will persist, and its value will compound. VivoScan™ positions C. elegans not as a legacy model, but as a modern, scalable, and AI-compatible platform for whole-organism insight, bridging the gap between computation and biology. By embedding organism-level evidence upstream, AI-driven pipelines can become not only faster, but smarter, safer, and more predictive of real-world outcomes.  

In the era of AI-accelerated drug discovery, high-quality biology remains the limiting factor. The ultimate purpose of all AI in drug development is to predict the safety and efficacy of compounds in whole organisms. This requires training data, which can only come from real biological experiments, and model predictions must be validated at every stage of the development pipeline. VivoScan™ ensures that AI is trained, validated, and guided by data that reflects how compounds behave in living systems - at an earlier stage than any competing approach.  

References:

[1] “Applications of machine learning in drug discovery and development”, Vamathevan et al., 2019, Nature Reviews Drug Discovery 

[2] “Caenorhabditis elegans as a Model System to Study Human Neurodegenerative Disorders”, Roussos et al., 2023, MDPI Biomolecules 

[3] “Artificial intelligence in biologic drug discovery: A review of methodological evolution and therapeutic applications”, Tang et al., 2026, Acta Pharmaceutica Sinica B 

[4] “C. elegans in high-throughput drug discovery.”, O’Reilly et al., 2014, Drug Discovery Today