From In Silico to In Vivo: How AI-Driven Drug Discovery Can Be Validated with Zebrafish Models

The pharmaceutical industry has embraced AI as a solution to persistent challenges in drug discovery and development, where traditional approaches require over a decade and cost billions of dollars, with a staggering 90% failure rate. AI-driven drug discovery platforms are now generating unprecedented opportunities by integrating machine learning, deep learning, and generative AI to revolutionize every stage of the pipeline. 

67 AI-discovered molecules were in clinical trials by 2023. Data reveals a high success rate in Phase I, reaching 80-90%. As of December 2023, 21 out of 24 AI-discovered molecules had completed Phase I trials, above historical industry averages (40%-65%). 

AI drug discovery promises to be a revolution, but despite its transformative potential, a fundamental challenge remains: computational predictions must be validated in biological systems.

Why Pure AI Predictions Aren’t Enough: Advantages of In Vivo Validation

AI drug discovery platforms rely on deep learning, machine learning, and generative AI to analyze massive biological datasets, perform virtual screening, and propose novel targets and molecules. These technologies fuel innovation but are not without significant limitations:

• Data quality matters: AI models are only as good as the training data. Most biological databases display bias (publishing mainly positive results), inconsistent assay standards and data collection practices, and incomplete metadata, all of which can misguide models and reduce predictive reliability. 
• Biological complexity: Living organisms are more than the sum of their parts. Simple computational models do not capture the multidimensional nature of toxicity involving multi-organ interactions, or the intricate network of off-target effects underlying real drug responses.​
• Black box limitations: Often, AI provides predictions without explainable mechanisms, creating barriers for regulatory approval and scientific acceptance. Validation through experimentation remains essential to confirm both efficacy and safety.​
• Systemic responses: The prediction of absorption, distribution, metabolism, excretion, and toxicity (ADME-Tox) is limited for novel, structurally complex, or multitarget compounds.  

In silico does not guarantee success in vitro or in vivo. Zebrafish offer a whole-organism approach compatible with throughput screening for target validation, hit-to-lead optimization, and lead refining by assessing efficacy and toxicity profiles. Zebrafish helps validate and prioritize targets and compounds discovered by AI before moving to costly mammalian models. 

Zebrafish as a Bridge: Biological Advantages for In Vivo Validation

Zebrafish have revolutionized preclinical validation with properties that uniquely bridge the gap between computational prediction and mammalian testing:

• Genetic and physiological similarity: Approximately 70% of human genes have at least one zebrafish ortholog, about 82% of human disease-related genes are conserved, and organ systems (heart, liver, nervous system) show a remarkable functional analogy.
• Transparency and scalability: The transparent embryos of zebrafish allow direct visualization of morphological development, organ function, and tagged compound localization using advanced imaging tools. Hundreds of embryos per pair enable large-scale screening at low cost.​
• Rapid development cycle: Zebrafish organs mature within 5 days post-fertilization (dpf), compressing toxicology and efficacy studies into days and allowing for fast project iteration.​

• High-throughput compatibility: Embryos arrayed in multi-well plates allow automated drug administration, robotic imaging, and deep phenotyping, ideal for validating the large libraries of compounds generated by AI. 
• Ethical advantages and the 3Rs: According to European guidelines, zebrafish embryos younger than 5 dpf are not classified as experimental animals, aligning with the 3Rs (Replacement, Reduction, Refinement) by allowing the reduction of mammalian animal use without sacrificing biological relevance.​
• Cost and time reduction: Zebrafish models can shorten drug discovery timelines by up to 40% while reducing early-stage costs by 60%, an enormous advantage for biotech and pharma companies evaluating thousands of computationally generated candidates.​

Case Studies & Use Cases: AI Predictions Validated in Zebrafish Models

One of the most compelling demonstrations of AI-zebrafish integration involves phenotypic drug screening and early toxicity testing. 

Valderrama et al. developed a framework integrating in silico genomic screening, in vitro metabolic validation, and in vivo zebrafish behavioral testing to identify psychobiotic candidates. Computational analysis predicted bacterial strains capable of metabolizing specific prebiotics and producing neuroactive molecules. Zebrafish larvae exposed to bacterial supernatants showed altered stress-related gene expression and behavior, validating computational predictions in a living organism. 

Another example close to our heart is the work developed by our sister company, ZeCardio Therapeutics. They have closed a whole target discovery loop: 

1. They have developed different dilated cardiomyopathy models to extract transcriptomic data from their damaged hearts. 
2. Zebrafish  data, together with DCMm patient data, has been analyzed through the use of a custom knowledge graph and the application of Graph Machine Learning (GML) algorithms, leading to the proposal of 50 potential targets.
3. They have closed the loop by validating experimentally the targets in the same disease models originally used. 
4. Through this approach, they have selected 10 new targets (20% efficiency) to move towards drug discovery stages.

Zebrafish have allowed them to generate proprietary, relevant data and in vivo validation of AI-derived hypotheses. The whole approach took under 1 year. Importantly, a similar approach using rodents would have taken 3 years and cost 10 times more.

At ZeClinics, we have participated in several projects that integrate artificial intelligence with experimental validation, contributing to this virtuous ‘AI + experimental in vivo cycle’, either by generating data from newly developed disease models or by validating AI-derived targets across various disease contexts, including osteoporosis, epilepsy, and chronic kidney disease. Altogether, these collaborations have enabled our clients to significantly reduce both the cost and duration of their projects, while obtaining clinically relevant data. 

If you're interested in learning more about how zebrafish can fit into your preclinical pipeline, contact us!

References

Jarallah SJ, Almughem FA, Alhumaid NK, Fayez NA, Alradwan I, Alsulami KA, Tawfik EA, Alshehri AA. Artificial intelligence revolution in drug discovery: A paradigm shift in pharmaceutical innovation. Int J Pharm. 2025 Jul 25;680:125789. doi: 10.1016/j.ijpharm.2025.125789.

Jayatunga MKP, Ayers M, Bruens L, Jayanth D, Meier C. How successful are AI-discovered drugs in clinical trials? A first analysis and emerging lessons. Drug Discovery Today. 2024; 29 (6): 104009. doi:10.1016/j.drudis.2024.104009.

Patton EE, Zon LI, Langenau DM. Zebrafish disease models in drug discovery: from preclinical modelling to clinical trials. Nat Rev Drug Discov. 2021 Aug;20(8):611-628. doi: 10.1038/s41573-021-00210-8.

Valderrama B, Daly I, Gunnigle E, O'Riordan K, Chichlowski M & Banerjee S, Skowronski A, Pandey N, Cryan J,  Clarke G, Nagpal J. From in silico screening to in vivo validation in zebrafish - a framework for reeling in the right psychobiotics. Food & function. 2025; 16. doi:10.1039/d4fo03932g. 

Zhang J, Li H, Zhang Y, Huang J, Ren L, Zhang C, Zou Q, Zhang Y. Computational toxicology in drug discovery: applications of artificial intelligence in ADMET and toxicity prediction. Brief Bioinform. 2025 Aug 31;26(5):bbaf533. doi: 10.1093/bib/bbaf533.

By Miriam Martínez

Miriam is a Human Biologist with a strong background in neuropharmacology and a passion for bridging science and innovation. After earning a master’s degree in the Pharmaceutical and Biotech Industry, she completed her PhD in Biomedicine at Pompeu Fabra University (Barcelona), where her research focused on the behavioral analysis of animal models for neurophenotypical characterization. Following her doctoral studies, Miriam transitioned into the healthcare marketing and communication sector, where she played a key role in developing impactful marketing strategies and educational campaigns for leading pharmaceutical brands. She now leverages her scientific expertise, strategic thinking, and creative communication skills in her current role at ZeClinics.