Can We Identify Drug Safety Risks Before They Become Costly Failures?

“The dose makes the poison.”

— Paracelsus (1493–1541), widely regarded as the father of toxicology

More than five centuries later, this principle remains fundamental to drug safety research.

However, the challenge facing modern drug development has evolved. Today, researchers are not only asking whether a candidate molecule can achieve therapeutic effects—they are increasingly asking:

Can we identify potential safety risks earlier, before significant resources are invested?

Many promising drug candidates fail not because they lack efficacy, but because safety concerns emerge too late in development. Drug-induced liver injury, cardiotoxicity, nephrotoxicity and gastrointestinal toxicity continue to represent major challenges in preclinical and clinical development, contributing to programme delays, increased costs and late-stage attrition.

Earlier identification of toxicity risks provides researchers with opportunities to optimise compounds, improve candidate selection and make more informed development decisions.

To address this challenge, Daxiang Biotech introduces Oakuloid™ ToxPredictor, an AI-enabled organoid-on-chip platform designed to support early toxicity prediction and human-relevant safety assessment.

Moving Beyond Traditional Toxicity Evaluation

Traditional toxicity assessment approaches, including animal models, conventional two-dimensional cell cultures and historical toxicity datasets, have played important roles in pharmaceutical research.

However, these approaches also face limitations:

· Animal models may not fully capture human-specific physiological responses due to species differences.

· Conventional cell models often lack the complex architecture and functional characteristics of human tissues.

· Historical datasets may contain heterogeneous data quality and limited representation of novel chemical structures.

· Single toxicity endpoints may not fully reflect complex biological responses.

Meanwhile, artificial intelligence has demonstrated significant potential in predicting molecular properties and identifying toxicity patterns. However, computational prediction alone cannot fully replace functional biological validation.

The future of toxicity assessment requires an integrated approach—combining the scalability of AI with the biological relevance of human-based models.

Combining AI Prediction with Human Organoid-on-Chip Validation

At Daxiang Biotech, we believe that better preclinical decisions require more human-relevant evidence.

Oakuloid™ ToxPredictor integrates AI-powered toxicity prediction with human organoid-on-chip models, creating a dual-pathway approach for early safety assessment.

In this integrated workflow:

AI models identify potential toxicity risks and prioritise candidate compounds.

Human organoid-on-chip models provide functional validation in biologically relevant environments.

Together, these capabilities help researchers evaluate:

· Which candidates may present potential safety concerns?

· Which predicted risks require further experimental validation?

· Whether therapeutic efficacy can be achieved within an appropriate safety window?

· Which optimisation strategies may reduce toxicity risks?

· Which candidates should be prioritised or discontinued earlier?

By connecting computational intelligence with human-relevant biology, Oakuloid™ ToxPredictor aims to support more efficient and informed drug development decisions.

Four Human-Relevant Toxicity Assessment Modules

Oakuloid™ ToxPredictor currently includes four application modules covering key organ systems frequently associated with drug safety risks.

1. Hepatotoxicity Assessment

The liver plays a central role in drug metabolism and remains one of the most important organs in drug-induced toxicity evaluation.

Drug-induced liver injury (DILI) is a major contributor to clinical development failure, regulatory warnings and drug withdrawal.

The hepatotoxicity module supports:

· Drug-induced liver injury risk assessment

· Hepatocyte injury evaluation

· Metabolism-associated toxicity analysis

· Candidate prioritisation based on liver safety profiles

2. Cardiotoxicity Assessment

Cardiac safety remains a critical consideration throughout drug development, particularly for compounds associated with cardiac injury or electrophysiological risks.

The cardiotoxicity module supports:

· Cardiomyocyte toxicity evaluation

· Cardiac safety risk assessment

· Dose–response analysis

· Safety window evaluation

3. Nephrotoxicity Assessment

The kidney plays an essential role in drug elimination and exposure regulation, making renal toxicity a key factor in candidate evaluation.

The nephrotoxicity module supports:

· Renal toxicity risk assessment

· Kidney injury evaluation

· Drug clearance-related safety analysis

· Candidate ranking based on renal safety profiles

4. Gastrointestinal Toxicity Assessment

The gastrointestinal tract is not only important for drug absorption but also represents a major site for drug-related adverse effects.

The gastrointestinal toxicity module supports:

· Intestinal epithelial injury assessment

· Barrier function evaluation

· Absorption-related risk analysis

· Gastrointestinal safety evaluation

Demonstrating Predictive Performance Beyond Known Chemistry

To evaluate the predictive capability of Oakuloid™ ToxPredictor, the platform was assessed using independent hepatotoxicity datasets containing chemically diverse compounds, including molecules with limited structural similarity to training data.

On the TDC-96 dataset (57 hepatotoxic compounds and 39 non-toxic compounds), Oakuloid™ ToxPredictor achieved an AUROC of 0.730, demonstrating strong discrimination capability and outperforming multiple conventional machine learning and deep learning approaches.

On the GECI-48 dataset (34 hepatotoxic compounds and 14 non-toxic compounds), Oakuloid™ ToxPredictor achieved an AUROC of 0.740, showing consistent performance and cross-dataset generalisation capability.

These results suggest that integrating AI-based prediction with human-relevant model systems can provide valuable insights for early toxicity assessment, particularly when evaluating chemically novel compounds.

Towards More Comprehensive Safety Assessment

Drug toxicity is rarely limited to a single organ system.

A candidate molecule may simultaneously influence metabolism, clearance, cardiovascular safety and gastrointestinal function.

Therefore, Oakuloid™ ToxPredictor is designed not only to evaluate individual toxicity endpoints but also to support a broader understanding of multi-organ safety profiles.

By enabling earlier identification of potential risks, the platform helps researchers make more confident decisions throughout the drug discovery and preclinical development process.

Supporting Smarter Decisions in Drug Development

Oakuloid™ ToxPredictor is designed for applications including:

· Early-stage candidate screening

· Lead optimisation

· Preclinical safety evaluation

· Toxicity-guided compound selection

· Multi-organ safety assessment

· Pipeline risk re-evaluation

As pharmaceutical research continues to advance toward human-relevant models, artificial intelligence and New Approach Methodologies (NAMs), early toxicity prediction will become increasingly important for improving translation and reducing development uncertainty.

At Daxiang Biotech, we believe the future of drug discovery is not only about finding medicines that work—but also about identifying safer medicines earlier.

Oakuloid™ ToxPredictor brings together AI intelligence and human organoid-on-chip biology to help researchers see safety risks earlier and make better decisions sooner.