Applications of Artificial Intelligence, Machine Learning, and Big Data Analytics in Drug Discovery and Pharmacovigilance: Current Advances and Future Perspectives

In today’s world, drug discovery, drug safety and Pharmacovigilance are fundamental pillar of modern healthcare, ensuring patient safety. However conventional drug discovery is time- consuming, highly costly and inefficient with high attrition rates during preclinical and clinical development. Traditionally, pharmacovigilance mainly relies on statistical signal detection manual case processing, spontaneous, reporting which often lead to delay ADR identification. But as technology grows Artificial Intelligence (AI), Machine Learning (ML), and Big Data analysis have emerged as transformation tools in pharmaceutical research. AI-driven algorithm can process massive chemical, biological and clinical database to predict molecular activity, toxicity and drug – drug interaction. In Pharmacovigilance Machine Learning (ML) model technologies enhances automated adverse data, signal prioritization and real-world evidences integration improving both speed and acceptance Furthermore, AI-based approaches have demonstrated significant potential in ADMET prediction and drug repurposing. Deep learning models, decision- tree algorithms, and knowledge -graph –based methods have improved ADR prediction and signal detection performance. This review ensures to summarizes recent development in Artificial Intelligence (AI), Machine Learning (ML) and Big Data application in drug discovery and pharmacovigilance highlights the strength and limitation of current approaches and discusses future perspective for integration these technologies into pharmaceutical research and safety monitoring.

2. Artificial Intelligence, Machine Learning and Big Data in Drug Discovery

2.1 Overview

Artificial intelligence, machine learning, and big data analytics have significantly transformed the drug discovery pipeline by enabling data-driven decision making at every stage. From target identification to lead optimization and drug repurposing, AI-based models reduce experimental cost, time, and attrition rates compared to conventional trial-and-error approaches

2.2 Target Identification: Target identification is the first step and most critical step in drug discovery, involving the selection of biological molecules that can modulate to treat disease.  With advancement in Artificial Intelligence (AI), Machine Learning (ML) methods such as Graph Neural Network (GNNs), Random Forest classifier (RF), Deep Learning models have been showed potential to find drug target with accuracy (Smith et al.). Such model are primarily used to analyze Genomics and Proteomics of target. Compared to traditional wet-lab screening, AI-based target identification significantly reduces experimental cost and time by prioritizing biological relevant targets from high-dimensional data.

2.3 Virtual Screening and Lead Discovery

Virtual screening is a key step in modern drug discovery that enables the rapid evaluation of large chemical libraries to identify potential lead compounds. Conventional virtual screening approaches primarily rely on molecular docking and scoring functions; however, these methods are often computationally intensive and associated with high false-positive rates. AI-based virtual screening employs supervised and unsupervised learning algorithms to predict ligand–target interactions, binding affinity, and structural compatibility. Deep learning models, particularly convolutional neural networks (CNNs), have been widely applied to analyze molecular structures and protein–ligand interaction patterns, leading to improved hit identification. Hybrid approaches that combine molecular docking with ML-based scoring functions have demonstrated superior performance compared to traditional docking methods alone. These hybrid models reduce false-positive predictions and prioritize compounds with higher biological relevance, thereby improving the overall hit-to-lead conversion rate. Furthermore, AI-driven virtual screening allows high-throughput analysis of millions of compounds within a significantly reduced time frame, accelerating early-stage drug discovery. Despite these advantages, AI-based virtual screening approaches face certain limitations, including dependence on high-quality training data and limited generalizability to novel chemical scaffolds. Therefore, experimental validation remains essential to confirm the biological activity of AI-predicted lead candidates. Overall, AI-enabled virtual screening represents a powerful and cost-effective strategy for lead discovery, complementing traditional experimental and computational methods.

2.4 Lead Optimization

Lead optimization is a critical phase of drug discovery that focuses on improving the potency, selectivity, safety, and pharmacokinetic properties of lead compounds. Traditional lead optimization relies on iterative chemical synthesis and biological testing, which is time-consuming and resource-intensive. The application of artificial intelligence (AI) and machine learning (ML) techniques has significantly improved the efficiency of this process by enabling data-driven optimization of molecular structures. AI-driven lead optimization utilizes deep learning models, decision-tree-based algorithms, and reinforcement learning approaches to predict structure–activity relationships and guide rational chemical modifications. These models can simultaneously optimize multiple parameters, including target affinity, selectivity, metabolic stability, and toxicity, thereby reducing the number of experimental iterations required. Generative models such as variational autoencoders (VAEs) and generative adversarial networks (GANs) have further expanded the capabilities of AI-based lead optimization by enabling scaffold hopping and exploration of novel chemical space. By predicting the impact of structural changes on biological activity, AI-based methods accelerate the identification of optimized lead candidates and reduce late-stage drug development failures. Despite these advantages, AI-based lead optimization approaches face limitations related to data availability, model interpretability, and synthetic feasibility of generated compounds. Consequently, close integration of AI predictions with medicinal chemistry expertise and experimental validation remains essential. Overall, AI-enabled lead optimization represents a powerful complementary strategy to traditional approaches, enhancing efficiency and decision-making in early-stage drug discovery.

2.5 ADMET Prediction

ADMET prediction, encompassing absorption, distribution, metabolism, excretion, and toxicity, is a crucial component of drug discovery aimed at minimizing late-stage drug development failures. Conventional ADMET evaluation relies heavily on in vitro and in vivo experiments, which are costly, time-consuming, and often conducted at advanced stages of development. The integration of artificial intelligence (AI) and machine learning (ML) techniques has enabled early and accurate prediction of ADMET properties, thereby reducing drug attrition rates. Deep learning models, including convolutional neural networks (CNNs) and recurrent neural networks (RNNs), have been widely applied to predict ADMET profiles using molecular fingerprints, physicochemical descriptors, and sequential molecular representations such as SMILES strings. These models demonstrate high predictive accuracy in assessing toxicity, metabolic stability, solubility, and drug–drug interaction potential. Ensemble learning approaches further enhance prediction robustness by combining outputs from multiple models. AI-driven ADMET prediction facilitates the early identification of compounds with unfavorable safety or pharmacokinetic characteristics, allowing their elimination before costly experimental validation. Despite these advantages, challenges remain in the application of AI-based ADMET prediction, including limited availability of high-quality annotated datasets, model interpretability, and regulatory acceptance. Consequently, AI-driven ADMET models are best utilized as decision-support tools rather than replacements for experimental evaluation. Continued integration of AI approaches with experimental pharmacokinetic and toxicological studies is essential for reliable drug development outcomes.

2.6 Drug Repurposing

Drug repurposing, also known as drug repositioning, involves identifying new therapeutic indications for existing or approved drugs. This approach offers significant advantages over traditional drug discovery, including reduced development time, lower costs, and established safety profiles. In recent years, artificial intelligence (AI), machine learning (ML), and big data analytics have emerged as powerful tools for accelerating drug repurposing by integrating diverse biomedical datasets and uncovering hidden drug–disease relationships AI-based drug repurposing leverages multiple data sources, including drug–target interaction databases, gene expression profiles, disease networks, electronic health records, and real-world data. Machine learning techniques such as network-based models, matrix factorization, and graph neural networks enable the identification of novel associations between drugs, targets, and diseases. In addition, natural language processing (NLP) techniques facilitate large-scale literature mining to extract relevant information from scientific publications and clinical trial reports. During recent public health emergencies, such as the COVID-19 pandemic, AI-driven drug repurposing approaches played a crucial role in rapidly identifying candidate drugs for clinical evaluation. Similar strategies have been applied in oncology, neurological disorders, and rare diseases, demonstrating the broad applicability of AI-based repurposing frameworks. By prioritizing repurposable candidates with high therapeutic potential, AI approaches improve decision-making and reduce experimental burden. Despite its promise, AI-based drug repurposing faces challenges related to data heterogeneity, algorithmic bias, and the need for experimental and clinical validation. Moreover, predicted drug–disease associations may not always translate into clinical efficacy Therefore, AI-driven drug repurposing should be considered a complementary strategy that supports hypothesis generation and prioritization rather than a standalone solution. Overall, the integration of AI and big data analytics has significantly enhanced the efficiency and scope of drug repurposing efforts in modern pharmaceutical research.

Table 1. Applications of Artificial Intelligence and Machine Learning in Drug Discovery