The pivotal role of artificial intelligence in enhancing experimental animal model research: A machine learning perspective

Artificial intelligence (AI) refers to a computer imitating “intellectual processes characteristic of humans, such as the ability to reason, discover meaning, generalize, or learn from past experience.”[1] Machine learning (ML) is a core area of AI to create predictive models by learning from data and gradually enhancing the capacity for prediction through experience.[2] The integration of AI and ML into animal research has shown promising potential to enhance translation and reproducibility, complementing traditional approaches such as animal models. AI and ML can optimize preclinical studies using animal models by analyzing complex datasets, improving experimental design, and predicting outcomes. This integration enables researchers to extract more meaningful information from animal experiments.[3] Combining AI/ML analyses of animal model data with human clinical data allows for better translation of findings. This integrated approach helps bridge the gap between preclinical and clinical studies, increasing the relevance of animal model findings to human disease. A combination of transcriptomic analysis (studying gene expression patterns) in postmortem human brain tissue from Alzheimer’s disease patients and mouse models of Alzheimer’s disease was done with the help of ML to identify dysregulated pathways associated with excitatory neurotransmission, a process crucial for brain function.[4]

This can contribute to the standardization of experimental protocols and data analysis in animal studies, promoting reproducibility. Automated analysis tools help reduce variability in results and enhance the reliability of findings across different laboratories.[5] Using ML algorithms to optimize drug dosing in mouse models of epilepsy was able to identify dosing regimens that reduced seizure frequency and severity while minimizing potential toxicities.[6] Integrating big data from both animal models and human studies using AI/ML allows researchers to identify commonalities and differences across species.[7] AI is employed to automatically track and analyze animal movements in their natural or controlled environments. This helps to understand behavior patterns, social interactions, and responses to environmental changes.[8] Various platforms, including MoSeq, DeepHL, DeepPoseKit, SLEAP, and DeepLabCut, employ deep learning techniques for animal behavior pose estimation. DeepLabCut, particularly, stands out as a widely utilized deep learning platform for behavioral analysis, addressing numerous limitations in animal posture and providing robust behavior tracking across diverse environments.[9] ML algorithms can be trained to recognize individual animals based on their facial features or unique markings. This is particularly useful in studying social structures and interactions within groups.[10] It also accelerates drug discovery by predicting the potential efficacy and side effects of pharmaceutical compounds.[11] Advancements in microfluidics-supported chemical synthesis, biological testing, and the integration of AI systems for iterative design improvement are laying the foundation for increased automation in these processes.[12] The creation of alcohol-preferring and apomorphine-susceptible rat strains through selective breeding aimed to mimic alcohol use disorder and schizophrenia. These phenotypes, influenced by genetic factors, necessitate a thorough behavioral analysis. ML can enhance this method by pinpointing animals with the most pertinent behaviors, streamlining the process before embarking on the extensive task of breeding multiple generations of animals.[13] In studies involving animals that use symbolic language, ML can help decipher and understand the meaning behind their signals.[14]

Bayesian ML models trained on high-throughput screening data indicated the potential repurposing of nicardipine or similar dihydropyridine calcium channel inhibitors for the treatment of Pitt–Hopkins syndrome, a rare hereditary illness presenting features of autism spectrum disorders (ASDs).[15] ML algorithms have been used to predict the functional consequences of differences in voltage-gated calcium as well as sodium ion channels, which have been connected to developmental encephalopathy, schizophrenia, and ASD.[16,17]

The application of AI in toxicity prediction aligns with advancements in data availability and algorithm capabilities.[18] Integrated AI approaches hold promise in transforming toxicology by predicting hazards for new chemical entities and reducing reliance on animal testing. Early rule-based expert systems evolved into statistical and machine-learning models such as Quantitative Structure Activity Relationship (QSAR).[19] The current era embraces deep learning, utilizing neural networks for toxicity predictions. One of the tools DeepTox standardizes the chemical representations of compounds, followed by the computation of numerous chemical descriptors utilized as input for ML methods. Subsequently, DeepTox undergoes training, evaluation, and the assembly of the most effective models into ensembles. Ultimately, the pipeline predicts the toxicity of new compounds.[20] Deep learning framework simultaneously models toxicity using in vitro, in vivo, and clinical data. Pretrained SMILES embeddings and Morgan fingerprints are the two distinct molecular-input representations. The multitask deep learning model demonstrates high accuracy in predicting toxicity across various endpoints, including clinical toxicity, as evidenced by robust performance metrics. More specifically, compared to current models in the MoleculeNet benchmark, the use of pretrained molecular SMILES embeddings improves clinical toxicity predictions.[21] A novel hybrid neural network (HNN), named HNN-Tox, is introduced for predicting chemical toxicity at various doses. This innovative approach combines two neural network frameworks, namely the convolutional neural network and the multilayer perceptron-type feed-forward neural network.[22] Another integration of eToxPred into protocols allows for the creation of customized libraries for virtual screening. This facilitates the exclusion of drug candidates that may pose potential toxicity risks or prove challenging to synthesize.[23]

The combined use of AI/ML and animal models also addresses various challenges. The challenges include working together across different fields and organizations to gather complete datasets that include clinical, neuroimaging, genetic, and biochemical information from both animal models and human groups. This collaborative approach will facilitate the development of robust AI/ML models capable of extracting meaningful insights from diverse data sources.[24] As AI/ML technologies continue to advance, it is crucial to prioritize ethical considerations in the use of these tools, particularly when working with sensitive data. Ensuring privacy and responsible data handling practices are essential for maintaining public trust and the integrity of research endeavors.[25] AI/ML models developed using data from animal models should undergo thorough validation in clinical settings. Validation studies involving diverse human populations and real-world clinical conditions are essential to ensure the reliability and generalizability of the models.[26]

In conclusion, the integration of AI and ML with animal model research holds immense potential for transformative discoveries. The field can maximize the impact of these technologies in advancing our understanding of preclinical studies and developing effective strategies for diagnosis and treatment.