A step towards replacing animal testing

In a recently published study, Dr Nature communicationResearchers have established AnimalGAN as a reliable alternative for generating synthetic pathology data to ultimately reduce animal testing in drug safety evaluation and accurately predict hepatotoxicity.

Study: A generative adversarial network model alternative to animal studies for evaluating clinical pathology. Image credit: Marques/Shutterstock.com

What is AnimalGAN?

To replace, reduce, and refine the 3Rs under the United States Food and Drug Administration (FDA) Modernization Act 2.0, the Animal Generative Adversarial Network (AnimalGan) emerges as a GAN model that generates clinical pathology data, challenging the ethical concerns of animal testing. by doing AnimalGAN outperforms quantitative structure-activity relationship (QSAR) in hepatotoxicity prediction and is similar to real animal studies.

By facilitating extensive virtual experiments, AnimalGAN can improve the prediction of rare toxic events and improve the translation of results from animals to humans. However, further research is needed to improve its predictive accuracy and solidify the role of AnimalGAN as a reliable alternative to animal testing.

About the study

The AnimalGAN initiative has helped the Toxicogenomics Project-Genomics Assisted Toxicity Evaluation Systems (TG-GATEs) database to advance predictive toxicology. The AnimalGAN model combines molecular descriptors and treatment conditions to simulate clinical pathology outcomes using a GAN framework, thereby integrating conditional GAN ​​(cGAN) with Wasserstein-GAN (WGAN) to increase stability and address small sample sizes.

The generator, G, obtained a molecular structure represented by an 1826-dimensional vector, dose level, treatment duration, and an 1828-dimensional random noise vector. The architecture of G was a fully connected network with five layers that generated a vector of clinical pathology indicators. The discriminator, D, evaluated these indices against treatment conditions and was constructed as a seven-level perceptron with dropouts to prevent overfitting.

AnimalGAN was trained on data from 8,078 mice, with 80% for training and 20% for testing. This model aims to replicate clinical measurements using metrics such as valid blood cell counts, cosine similarity, and root mean square error (RMSE) for validation.

After 6,000 epochs, the data generated by AnimalGAN closely reflects real data. In addition, AnimalGAN’s performance was tested against unseen data, thus confirming the predictive ability of the model. AnimalGAN predictions were also benchmarked against QSAR predictions, showing differences in predictive performance.

For toxicity assessment, AnimalGAN outputs were compared with actual experimental results to confirm its consistency. External validation with the Drugmatrix dataset confirmed the viability and applicability of the model, thereby indicating its potential as an alternative to animal testing in predicting clinical outcomes.

Study results

AnimalGAN, a new model in computational toxicology, has demonstrated impressive capabilities by generating 38 clinical pathology metrics and simulating complex biological responses to different treatment lengths and doses. AnimalGAN was thoroughly trained on data from 6,442 mice across 1,317 distinct treatment conditions with 110 compounds from the TG-Get database.

AnimalGAN’s performance was evaluated against a new cohort of 1,636 mice. The results show a striking match between the synthetic data produced by AnimalGAN and the real clinical data, which was highlighted by a low error margin and perfect matching of pattern matching. The use of T-SNE for visual confirmation further underscores the model’s accuracy in simulating real-world biological outcomes.

AnimalGAN’s power was rigorously evaluated using three challenging scenarios, each designed to test the model’s ability to reliably predict a variety of drug outcomes. The trials involved drugs with significantly different chemical structures, therapeutic categories, and timing of FDA approval than those used to manufacture AnimalGan. Remarkably, the model consistently replicated its initial success, thus showing its reliability, even when applying drugs that were distinctly different from its training set.

AnimalGAN’s performance was compared to conventional artificial intelligence (AI) methods, such as quantitative structure-activity relationship (QSAR) models, which are typically modeled to predict each clinical pathology measure separately. In comparison, AnimalGAN was associated with an impressive ability to predict all 38 measurements simultaneously with greater accuracy, thus highlighting its improved predictive ability over traditional models.

The real-world applicability of AnimalGAN was confirmed in a typical toxicological assessment scenario, where the model was tasked with comparing treatment groups with control groups to establish safety margins. Model predictions closely aligned with actual animal test data and achieved near-perfect agreement rates. This highlighted the potential of AnimalGAN as a powerful tool for evaluating hepatotoxicity and nephrotoxicity, thus suggesting that it could significantly reduce the need for animal testing in this area.

An external validation of AnimalGAN using data from the DrugMatrix database was conducted to further assess its accuracy. Despite the inherent variability of clinical pathology measurements across different experimental settings, AnimalGAN achieved greater than 80% consistency when comparing results between datasets, thus strengthening its applicability and reliability in different situations.

AnimalGAN also predicts the risk of idiosyncratic drug-induced liver injury (IDILI), a daunting challenge in drug safety monitoring. By virtually replicating the clinical pathology of a large rodent population, AnimalGAN was able to predict the probability of iDILI occurrence. Furthermore, the model distinguished the risks associated with a set of diabetes drugs, thus confirming its valuable contribution to the identification of potential drug safety problems before they emerge in clinical settings.