IN-SILICO TOXICITY PREDICTION TOOLS: A REVIEW OF TECHNIQUES AND APPLICATIONS

IN-SILICO TOXICITY PREDICTION TOOLS: A REVIEW OF TECHNIQUES AND APPLICATIONS

Shambhavi Shahi * and Sanjay Sharma

Shobhaben Pratapbhai Patel School of Pharmacy & Technology Management, Mumbai, Maharashtra, India.

ABSTRACT: A vital tool in drug discovery and development, in-silico toxicity prediction tools and algorithms provide economical, time-efficient and morally sound substitutes for traditional in-vitro and in-vivo toxicological testing. Computational techniques are essential to predict the possible toxicity and side effects prior to clinical testing because of the ever-increasing complexity of pharmaceutical compounds and the regulatory bodies' increasing focus on early toxicity screening. An overview of the in-silico methods for toxicity prediction is discussed in this article, which includes pharmacophore modeling, machine learning algorithms, molecular docking, and Quantitative Structure–Activity Relationship (QSAR) models. It weighs the underlying principles, prediction powers and practical applications of popular database tools including ProTox-II, SwissADME, Toxtree, and ADMETlab. The review also includes case studies that show how the tools are used to evaluate the toxicity of potential drugs. Limitations including model generalizability, lack of standardization, and validation issues persist despite the tool’s increasing dependability. The final section of the analysis looks at the state of regulations today and suggests ways to improve the accuracy and acceptability of in-silico toxicity forecasts in the pharmaceutical sector.

Keywords: In-silico toxicity, QSAR models, Drug safety, ProTox-II, ADMET prediction, SwissADME, Computational toxicology

INTRODUCTION: In order to increase the effectiveness and dependability of early drug research and development operations, the pharmaceutical industry’s dependability is inclined towards computational tools. Among these, in-silico toxicity prediction has emerged as a crucial technique to assess potential adverse effects of drug candidates without the need for extensive animal research or early-stage clinical trials.

Because of the rising costs and ethical dilemma associated with in-vivo and in-vitro toxicological studies, the shift to computer-aided toxicity screening is consistent with the global drive for the 3Rs principle Reduction, Refinement, and Replacement of Animal Testing. Toxicology is the leading cause of medication attrition during clinical advancement, and it plays a major role in the failure of Phase II and Phase III trials.

Therefore, in addition to reducing research costs, early identification of harmful liabilities is vital for guaranteeing patient safety and regulatory compliance. Investigators can forecast toxicological outcomes like hepatotoxicity, cardiotoxicity, carcinogenicity, and mutagenicity using in-silico approaches, which take into account molecular structure, physicochemical properties, and biological interactions. These predictions are basically supported by a range of computational methods including molecular docking, artificial neural networks, machine learning, pharmacophore mapping, and Quantitative Structure–Activity Relationship (QSAR) modeling. The development and availability of open-source and commercial in-silico tools such as ProTox-II, SwissADME, Toxtree, and ADMETlab have made it possible for even early-stage researchers to review compounds for safety profiles with a reasonable degree of confidence.

These techniques are being incorporated more and more into academic research, regulatory agencies, and the pharmaceutical industry's decision-making processes. In spite of their increasing usefulness, in-silico models have drawbacks. External validation is still difficult, as model prediction frequently depends on the calibre and variety of training datasets. Furthermore, the lack of a consistent regulatory framework prevents these models from being widely accepted as independent tools for decision-making.

This study discusses the fundamental methods, widely used software and platforms, practical applications and regulatory considerations in order to give readers a thorough grasp of in-silico toxicity prediction tools. Additionally, it discusses existing drawbacks and suggests ways to improve these tools' predictive accuracy in pharmaceutical research in the future.

Need for Toxicity Prediction in Drug Development: An important aspect as to why pharmaceutical research and development fails is still toxicity. Unexpected toxicological consequences found at the preclinical or clinical stage account for about 30% of drug candidates' failures ¹. Therefore, early toxicity screening is essential to lowering development costs, minimizing risk and enhancing candidate selection ². Toxicology must be included into the drug development cycle to identify and rule out high-risk candidates ahead in time in order to reduce these hazards ³. Conventional toxicity assessment techniques, like in-vitro cell-based tests and in-vivo animal studies, are costly, time-consuming, and frequently produce data with little application to humans. The search for alternate approaches has also been fueled by growing regulatory pressure and ethical considerations about animal welfare ⁴.

Using computer models acquainted with past toxicological data, in-silico toxicity prediction enables quick evaluation of sizable chemical libraries. From lead optimization to regulatory submission, these tools can be used at any point in the process, catering to drug development principles which are sustainable and ethical which enables well-informed decision-making ⁵

Overview of In-silico Approaches: In-silico models use diverse computational technique to simulate biological responses or infer toxicity potential based on molecular property of the compound. Few common approaches include:

Quantitative Structure–Activity Relationship (QSAR): One reliable in-silico method that links the chemical structures of molecules to their biological activity is the Quantitative Structure-Activity Relationship ⁶. There is a mathematical correlation between known toxicological endpoints and molecular descriptors like hydrophobicity, molecular weight, and electronic characteristics. The QSAR models are frequently used for regulatory purposes, especially in REACH and OECD recommendations, and can be either linear or non-linear ⁷. Since in-silico approaches offer computational tools to speed up the identification and optimization of the drug candidates, they have become essential in today's drug development cycle ⁸. The basic idea in QSAR is that comparable compounds have similar activities, QSAR makes it possible to create predictive models that are based on the quantitative link between target activities and molecular structures ⁹. A training set of molecules with known activities is used to build these models, and the relationships that are obtained are then used to forecast the activities of new compounds ¹⁰. Early on in the drug development process, QSAR models are essential because this enables researchers to rank drug candidates according to their advantageous ADME/T characteristics, which expedites the selection process ¹¹. New molecular activities can be predicted using QSAR ¹². The two main categories of in-silico drug discovery techniques are structure-based drug discovery and ligand-based drug discovery. The selection of a representative and varied training set of molecules is the first of manyunique phases in the QSAR models. After that, molecular moeities are transformed into numerical descriptors that represent the molecule’s electronic, structural, and physical attributes ¹³. These descriptors might be three-dimensional (3D), which takes into account the atomic spatial arrangement, or two-dimensional (2D), which captures topological structures. Non-linear approaches like neural networks should be employed since complex biological processes do not exhibit a linear relationship ¹⁴. These models are constructed using a variety of statistical techniques, including machine learning algorithms, partial least squares regression, and multiple linear regression. To guarantee the model’s resilience and prediction precision, they undergo extensive internal and external validation procedures.

Molecular Docking: Molecular docking mimics how any chemical substance shows interaction with a biological target, like DNA, enzymes, or receptors. It forecasts a molecule's orientation and binding affinity, which may reveal possible off-target toxicity (e.g., cardiotoxicity via hERG binding). Molecular docking is essential for identifying and refining possible therapeutic options as a fundamental computational method in contemporary drug discovery. It evaluates the binding affinity of the interactions and then makes it easier to anticipate the binding orientations of small molecule ligands within the binding site of a macromolecular target, like a protein or nucleic acid ¹⁵.

Early-stage drug discovery benefits highly from this computational method since it speeds up the drug identification process of viable drug leads by screening large chemical libraries to locate novel ligands that show complementarity to the target's binding pocket ¹⁴. Since precisely anticipating binding shape and affinities is still a difficult process, molecular docking is not without its difficulties ¹⁵. The molecular docking process is a two-step process ¹⁶. To produce a wide range of potential binding positions, a search algorithm first examines the ligand's conformational space inside the binding site ¹⁷. Second, each pose is assessed using a scoring formula that evaluates the ligand-target binding affinity ¹⁸.

Scoring functions usually consider phenomena like hydrogen bond interactions, van der Waals forces, electrostatics, solvation, and entropy to quantify the strength of the intersection ¹⁹. The quality of the protein structure the molecular docking study is based on influences its reliability ²⁰. While high-resolution X-ray crystal structures are the gold standard, cryo-electron microscopy-derived homology structures are acceptable in their absence. The use of molecular docking, molecular dynamics, and free energy perturbation in tandem facilitates more accurate calculations of binding affinity and improves binding pose predictions ²¹. The ability to perform in silico molecular docking studies has revolutionized the processes involved in the design of new drugs by providing realistic timelines and enabling the rapid evaluation of numerous ligands, which simplifies the comparison of scores ²².

Pharmacophore Modeling: Pharmacophore modelling method identifies the spatial arrangement of criteria needed for the functions of a living organism (biological functions). Criteria such as hydrophobic groups and hydrogen bond donors are included ²³. In pharmaceutical Research and Development (R&D), pharmacophore modelling served as a means to identify the essential structural features of the molecules that can elicit a certain biological activity. They abstract the functions which are essential for molecular interactions by portraying them as features which are not specific to a certain chemical structure ²⁴.

From groups of active chemicals or through observable interactions between proteins and ligands, pharmacophore models may be created. Such interactions can be captured in X-ray crystal structures, NMR structures, or even docking poses ²⁴. The rationale of pharmacophore models relying on known active ligands is that those ligands are presumed to interact with a common receptor ²⁵. The creation of ligand-based pharmacophore models involves algorithms designed to create many ligand conformers which are close to the bioactive conformation, capturing important spatial patterns of the chemical features ²³. This approach is advantageous when the 3D structure of the target protein is not known. For activities like virtual screening, lead identification and lead optimization, the pharmacophore modelling becomes essential ²⁶.

Pharmacophore models are frequently employed to virtually sift through vast chemical libraries to identify molecules that are likely to interact with a given biological target ²⁷. The determination and refinement of pharmacophore features is more accurate if a protein-ligand complex structure is available. The three-dimensional (3D) dimensional data of the ligand is important in its identification and placement of pharmacophore features that correspond to the functional groups that will interact with the protein. To reduce costs and speed up the processes in the drug development cycle, the models serve to identify novel chemical entities which are likely to alter the targets of the disease ²⁸.

Furthermore, the technique applies to any naturally occurring system in which biological activity involves receptors and ligands ²⁹. Pharmacophore models serve greatly in the area of lead optimization through the thoughtful creation of more potent and selective analogs with better pharmacokinetic properties enabling their rational design ³⁰. These models assist medicinal chemists in relating structure-activity relationships and allow the designs of complex molecules with preferred binding affinities and desired biological actions ³¹.

Machine Learning and AI-Based Models: The R&D sector of pharma will surely feel the impact of AI in the near future, given how it is transforming everything else. In-silico toxicology has AI advancements working in its favor. Toxicological datasets are utilized for training supervised learning algorithms like random forests, support vector machines, and deep neural nets which predict or classify toxicity ³². These models are far superior to the traditional QSAR models. AI and machine learning in modern times has significantly bolstered and even transformed, numerous functions of pharmaceutical R&D which include drug discovery, optimization, and even repurposing, owing to the sophisticated in-silico tools available today ³³. These computational models overcome the limitations of conventional drug discovery techniques. These models address the concerns associated with steep costs, prolonged timelines, and high failure rates of conventional methods by sophisticated algorithms, modeling intricate biological systems, predicting drug-target interactions, and rapidly identifying promising drug candidates from extensive datasets ^{34, 35}.

In-silico methods include molecular dynamics simulations, network analysis, pharmacophore and homology modeling, quantitative structure-activity relationships (QSAR) and other computational methods. The predictive models of pharmacologic processes which are constructed using these algorithms, are a result of combining these methods with machine learning ³⁶. The efficacy of new age drug discovery has greatly benefitted from the incorporation of computational techniques, which have exponentially accelerated and enhanced the precision of these processes ³⁷. The incorporation of computational techniques has emerged as a crucial element of drug development. Machine learning algorithms, such as deep learning neural networks, are capable of analyzing biomedical data on a vast scale which enables them to uncover buried connections between drugs and diseases. This capability enables the identification of new therapeutic uses for old drugs, which accelerates their translation into clinical use, also referred to as repurposing ³⁸.

Using AI algorithms to search for new therapeutic targets and potential drug candidates in vast databases has proven to be far better than the traditional trial-and-error approach. Analysis of deep databases to find complex patterns allow AI to make intricate mappings and ³⁹.

From focusing on improving productivity through computer-aided drug design in the 1960s and 1970s, the field of AI in drug discovery has managed to evolve to sophisticated machine learning algorithms today ⁴⁰. The vast amounts of genomic and transcriptomic data on diseases, healthy tissues, and cell lines has made it possible to create the AI-based software and tools needed, which researchers can now freely access and apply in their work ⁴¹.  AI-based models are applied in multiple stages of the drug discovery and development processes, starting from target identification and diagnosis to screening and lead discovery ⁴². Just as reinforcement learning optimizes certain chemical properties, generative models aid in the creation of new synthetic compounds ⁴³. AI plays a vital role in streamlining patient selection, real-time monitoring, tailoring reactions to enhance trial efficiency, and optimizing overall procedural efficacy ⁴⁴. AI algorithms that search through vast chemical libraries for potential candidates streamline the process, improving the speed and accuracy of the search. In the clinical environment, AI has significantly enhanced tailored medicine by improving treatment selection, minimizing side effects, and fine-tuning dosage ⁴⁵. The use of AI in drug development has the potential to enhance the precision of predicting a medicine's safety and efficacy, which would raise the pipeline's success rate while also cutting costs and shortening the time on the development cycle ⁴⁶.

Commonly Used Tools and Databases: Multiple tools have been developed and upgraded to predict toxicity using in-silico techniques. These platforms integrate databases, algorithms and decision rules. Key tools include:

ProTox-II: ProTox-II is an online tool for forecasting toxicological endpoints which includes mutagenicity, carcinogenicity, immunotoxicity, and hepatotoxicity as well as oral toxicity. It brings into use of the machine learning models, fragment propensities and chemical similarities. In the field of pharmaceutical R&D, ProTox-II is a crucial in-silico tool that helps researchers forecast the toxicity of potential drugs early in the development process. Utilizing computational power, in-silico techniques have grown as a pillar in pharmacology wherein they support the development, testing, and improvement of drug discovery procedures ⁴⁷. Proteomics and cheminformatics, which uses informatics techniques to fix chemical problems, must be combined in order to extract useful information from the vast amounts of data produced by evaluating drug candidates across a variety of biological sites and phenotypic screens ⁴⁸.

The use of computational techniques is consistent with the present trend, promoted and accepted by agencies such as the European Chemicals Agency and the Organization for Economic Co-operation and Development, of substituting in-vitro or in-silico models for non-clinical studies ⁴⁹. Compared to conventional animal testing, these models are less expensive and time-consuming while improving the effectiveness and moral implications of medication development ⁵⁰⁵¹. Early in-silico predictions are extremely important since toxicity evaluation is a critical bottleneck in drug development that frequently explains why therapeutic candidates fail in later stages ⁵². By analyzing chemical or biological trends this tool can create computational models that use data from earlier efforts to quickly identify compound toxicity ^{53, 54}.

Toxtree: Toxtree is an open-source program that uses decision tree-based criteria to detect toxicological risks, which includes skin sensitization, genotoxic carcinogenicity, and Cramer categorization. It is frequently utilized in terms of regulatory context. Computational techniques have become essential tools in the field of pharmaceutical development for forecasting the probable toxicity of medication candidates as well as other attributes ⁵⁵. Early in the drug development process, researchers can prioritize molecules with acceptable safety profiles by using Toxtree which provide a quick and affordable substitute for conventional in-vivo and in vitro approaches ^{56, 57}. It also provides a flexible platform for evaluating the possible negative effects of chemicals and is essential to pharmaceutical R&D ⁵⁸.

Toxtree uses a decision tree technique that makes use of well-established toxicological laws and knowledge sources. Toxtree's strength is its capacity to forecast toxicity based on chemical structure ⁵⁹. When working with novel substances, when experimental toxicity data is few or nonexistent, this capability is quite helpful. A free and open-source program called Toxtree uses a decision tree method based on the chemical structure of the substance being studied to help anticipate toxicological dangers ⁵⁰. Users can choose particular prediction techniques that are pertinent to their study issue thanks to the software's modular design, which covers a broad range of endpoints such mutagenicity, carcinogenicity, skin sensitization, and developmental toxicity.

Toxtree can highlight possible hazardous effects by examining the molecular structure and detecting the presence of structural warnings. This provides useful information for compound prioritization and directing structural changes to reduce toxicity risks. Researchers with different degrees of computational ability can use Toxtree because of its intuitive interface and thorough documentation. Additionally, because it is open-source, it encourages transparency and permits community-driven creation and verification of novel prediction techniques. Toxtree's incorporation into pharmaceutical processes can minimize the chance of late-stage failures brought on by unanticipated toxicities while also greatly speeding up the identification of viable therapeutic candidates ⁶⁰.
To forecast the ADMET (absorption, distribution, metabolism, excretion, and toxicity) characteristics of drug candidates, the pharmaceutical industry uses in-silico techniques, such as programs like Toxtree ⁶¹.

The four main processes that affect a drug's destiny in the body are absorption, distribution, metabolism, excretion, and toxicity, or ADMET. To find molecules with the best pharmacokinetic and safety profiles, it is essential to assess these characteristics early in the drug discovery process ⁶². By forecasting possible negative effects based on a compound's chemical structure, Toxtree aids in toxicity evaluation, a crucial part of ADMET profiling. From hit detection to lead optimization, Toxtree is used in pharmaceutical research at several phases of drug development. Researchers can concentrate their efforts on compounds with a higher chance of success by eliminating those with undesirable toxicity profiles by incorporating Toxtree into virtual screening methods. Toxtree and other in-silico techniques improve knowledge of a drug's effectiveness and adverse effects ⁶³. Toxtree can direct structural changes during lead optimization that are intended to lower toxicity while preserving or enhancing medication efficacy. A government agency partnership called Tox21 uses large datasets to identify connections between structure, function, and possible toxicity ^{64, 65}. Furthermore, Toxtree is essential for evaluating the safety of metabolites, excipients, and contaminants, which guarantees the general safety and caliber of pharmaceutical products ^{66, 67}.

SwissADME: SwissADME offers information on bioavailability, lipophilicity, and possible toxicophores that may be connected to toxicity profiles, even if its primary application is in pharmacokinetic prediction. With its extensive set of features for predicting the absorption, distribution, metabolism, and excretion properties of drug candidates, SwissADME is a crucial online tool in the pharmaceutical industry ⁶⁸. These ADME characteristics, which affect a drug's bioavailability, tissue distribution, metabolic stability, and elimination pathways, are essential factors in determining its efficacy and safety profile ⁶⁹. Early in the drug development pipeline, researchers can obtain important insights into the ADME features of their compounds because to SwissADME's prediction power, which is derived from its use of complex computational algorithms and large databases ⁶⁸.

Since ADME characteristics have historically been a major factor in medication failure during clinical trials, even beyond worries about effectiveness or safety, the ability to predict these qualities accurately is crucial ⁷⁰. Researchers can speed up development and lower the chance of late-stage failures by proactively identifying and mitigating potential ADME-related risks by incorporating SwissADME into the early phases of drug discovery ⁷¹. SwissADME can also predict a number of important physicochemical characteristics that affect a drug's ADME profile. These include water solubility, which determines a molecule's ability to dissolve in aqueous media and is crucial for absorption and distribution; topological polar surface area, which reflects the molecule's hydrogen bonding capacity and influences its membrane permeability; and lipophilicity, which measures a molecule's affinity for lipid environments and affects its ability to cross biological membranes ⁷². SwissADME also makes it easier to forecast how a molecule can interact with important drug-metabolizing enzymes, like cytochrome P450s, which are essential for xenobiotic detoxification and removal. Drug candidates with desired ADME/T qualities can be chosen through the use of in silico tests in the early phases of drug development ⁷³. Predicting possible drug-drug interactions and maximizing drug metabolism need an understanding of these interactions ⁷⁴.

It is possible to evaluate the efficacy and biopharmaceutical properties of possible drug candidates simultaneously by calculating how long a medication will last in the bloodstream. Researchers can prioritize compounds with good ADME features according to the tool's reasonable accuracy in predicting these traits, which lowers the likelihood of unexpected problems throughout preclinical and clinical development. SwissADME is used in many different pharmaceutical research and development projects. It is a useful tool for virtual screening in the field of drug development, allowing scientists to sort through sizable libraries of compounds according to their anticipated ADME characteristics and find intriguing candidates for additional research. By helping medicinal chemists create compounds with better ADME characteristics, it facilitates lead optimization ⁷⁵. The drug development process is streamlined and total expenses are decreased by SwissADME's quick and affordable ADME predictions, which enable researchers to make well-informed decisions on compound selection and optimization. Furthermore, SwissADME is essential to drug repurposing initiatives, which assess already-approved medications for their capacity to treat novel illnesses. The arithmetic average and p-value indices demonstrated a significant difference in the ADMET-score between the three data sets ⁷⁶. By forecasting the ADME features of already available medications, scientists might find potential candidates that might have advantageous traits for a novel therapeutic use.

ADMETlab and ADMET Predictor: More than 200 descriptors pertaining to absorption, distribution, metabolism, excretion, and toxicity are available through the web-based program ADMETlab. It predicts toxicity endpoints using a variety of algorithms and big training datasets. Predicting features such as absorption, distribution, metabolism, excretion, and toxicity is essential in pharmaceutical development to find viable drug candidates early in the process ⁷⁷. With computational approaches to assess these features, in-silico tools such as ADMETlab and ADMET Predictor have become invaluable resources in this attempt, speeding up the drug development process and decreasing the need for resource-intensive experimental methods ⁷⁸. As a complete platform, ADMETlab offers a broad range of ADMET predictions derived from quantitative structure-activity connection models and different machine learning techniques ⁷⁹. Researchers from a variety of backgrounds can use it because of its user-friendly interface and integration of several features, such as data curation, model construction, and result display. A number of essential modules, including data preparation, model training, prediction, and molecular descriptor computation, are commonly included in ADMETlab's design ⁸⁰. cheminformatics toolkits are used to create molecular descriptors, which encode the physicochemical and structural characteristics of substances. The machine learning algorithms use these descriptors as inputs after being trained on large datasets of substances with known ADMET profiles. A range of techniques, each with unique advantages and disadvantages, are used in the models built in ADMETlab, including support vector machines, random forests, and neural networks.

The commercial program ADMET Predictor, on the other hand, is well-known for its strong and precise ADMET predictions. It uses complex algorithms and large datasets to produce trustworthy estimates of drug-like characteristics. It sets itself apart by providing real-time feedback on how structural modifications impact molecular characteristics and dynamically updating predictions as structures are modified. This is particularly helpful when creating analogs of recognized leads ⁸¹.

A variety of sophisticated characteristics are included in ADMET Predictor, such as mechanistic models that take into consideration particular biological processes involved in ADMET, like transporter interactions and enzyme kinetics. Additionally, the platform facilitates the creation of bespoke models and the incorporation of proprietary data, allowing users to customize the forecasts to meet their own requirements ⁸². By employing deep learning models based on encoder and decoder architectures to predict interactions between medications and their targets, these platforms make a substantial contribution and offer alternatives to experimental methods ⁷⁹. These models forecast characteristics like toxicity and bioactivity by using the structural data represented in molecular graphs ⁸³. Both ADMETlab and ADMET Predictor are useful for a variety of tasks, such as medication repurposing, lead optimization, and virtual screening. Researchers can lower the chance of late-stage failures brought on by poor pharmacokinetics or toxicity by prioritizing drugs with favorable profiles by forecasting ADMET characteristics ⁸⁴. Additionally, these in-silico technologies aid in the detection of possible drug-drug interactions and direct the development of safer and more efficient pharmaceuticals. In the end, these platforms improve drug discovery's efficacy and efficiency by bridging the gap between experimental validation and computational predictions and assisting in the creation of innovative treatments ⁸⁵.

VEGA and ToxCast: Numerous predictive models are hosted on the VEGA platform, which also permits external validation. The U.S. EPA's ToxCast tool predicts chemical dangers by using high-throughput screening data from hundreds of experiments. The use of in-silico approaches has transformed drug discovery and safety evaluation in the field of pharmaceutical research and development ^{86, 87}. VEGA and ToxCast are notable resources among the many computational tools accessible, providing important information about the pharmacological and toxicological characteristics of chemical substances. These tools support the "3R" principle, which promotes moral and effective research practices, and are in line with the current focus on substituting in-vitro or in-silico tests for conventional

Because computational methodologies allow for the quick prediction of activity for a large variety of compounds during virtual screening exercises, public authorities and international organizations actively promote their usage. A complete platform for forecasting different physicochemical, toxicological, and environmental destiny features of chemical compounds is offered by the QSAR modeling package VEGA ⁸⁸. Its ability to quantitatively evaluate the connections between a molecule's structure and its wide variety of biological functions is its strongest point ⁸⁹. Early identification of possible drug candidates with advantageous pharmacokinetic and safety characteristics is made easier by VEGA, which uses complex algorithms and large chemical databases to estimate important parameters like drug absorption, distribution, metabolism, excretion, and toxicity profiles ⁸⁰. To create models based on the idea that related compounds have similar actions, the quantitative structure-activity relationship was created for VS ⁹⁰. A sizable chemical library is whittled down by QSAR models to a select few compounds for experimental testing as part of the screening process ⁹¹. Virtual screening in conjunction with virtual combinatorial chemistry has become a vital tool in drug discovery, allowing for significant cost savings and expedited procedures ⁹². One or more SARs can be captured and encoded using a number of models, which can subsequently be used to forecast the activities of novel compounds.

In contrast, the US Environmental Protection Agency launched the extensive toxicity testing program ToxCast, which uses high-throughput screening assays to assess the possible toxicity of thousands of chemicals ⁹³. ToxCast uses a battery of in vitro assays that span a wide range of biological pathways and cellular activities, in contrast to traditional animal testing methods. This allows for the quick and economical evaluation of chemical safety. The information produced by ToxCast is openly accessible and is a useful tool for comprehending chemical toxicity mechanisms and creating risk assessment prediction models for human health. The Tox21 initiative, which seeks to improve toxicity assessment techniques, provides the data used in ToxCast ⁶⁵.

Since QSAR models establish a mathematical connection between toxicity and chemical structure, they can also be used to evaluate the toxicity of untested substances ⁹⁴. QSAR models and in vitro techniques are typically used to support the predictions in the setting of read-across. Pharma researchers can obtain a more comprehensive grasp of the possible hazards and advantages connected to new drug candidates by utilizing the complimentary qualities of VEGA and ToxCast. For example, ToxCast data can be used to evaluate a compound's ability to cause particular toxicological effects, whereas VEGA can be used to forecast the compound's ADMET qualities. In addition to speeding up the drug development process, this integrated strategy improves the efficacy and safety of novel drugs.

TABLE 1: AVAILABILITY OF PREDITION TOOLS

Case Studies and Applications:

Early Screening of Antidiabetics: Off-target effects like cardiotoxicity or hepatotoxicity have been predicted using in-silico models for drugs like DPP-4 inhibitors. By eliminating compounds that have a high likelihood of becoming immunotoxic or hepatotoxic, tools like as ProTox-II and ADMETlab have been used to improve lead selection.

Toxicity Profiling of Herbal Extracts: To help create safer nutraceuticals, researchers are increasingly using QSAR and docking techniques to predict the hepatotoxic or nephrotoxic effects of herbal bioactives.

Regulatory Submission for REACH: When models adhere to OECD validation guidelines, QSAR predictions are permitted under the European REACH rule. Data for chemical dossiers is frequently generated using programs like VEGA and Toxtree.

Industry Applications: To lessen the burden of animal testing and improve the prediction of uncommon toxicities, including idiosyncratic liver injury, pharmaceutical companies have incorporated in-silico screening into their early R&D pipelines.

Strengths and Limitations of In-silico Models: The pharmaceutical business now relies heavily on in-silico models, which have several uses ranging from drug discovery to drug delivery system optimization ⁹⁵. Absorption, distribution, metabolism, excretion, and toxicity (ADMET) characteristics, drug-target interactions, and possible therapeutic efficacy are all aided by these computational techniques, which include molecular docking, molecular dynamics simulations, and quantitative structure activity relationship modelling ^{96, 97}. In-silico techniques have greatly sped up the lead identification process, saving time and money by quickly screening large chemical libraries and forecasting the activity of new compounds ⁹⁸. Additionally, these models make it easier to characterize the links between structure and activity. They can also be used to prioritize huge screening decks or even create novel compounds from scratch. Additionally, in-silico trials are being used to improve animal experiments and to virtually test pharmaceutical treatments ⁹⁹. Finding antibacterial and antivirulence drugs and comprehending their mechanisms are thought to be more affordable when in-vitro research are combined with in-silico techniques ¹⁰⁰. The use of computational methods is consistent with the "3R" principle's rising emphasis on substituting in vitro or in-silico tests for non-clinical testing. It's critical to recognize the limitations of in-silico technologies, even though they present a potential way to speed up drug development and cut costs ¹⁰¹.

In spite of all of its benefits, in-silico models have drawbacks. The accuracy of the models itself is a significant obstacle. The quality of the data and algorithms that these models are based on determines how good they are. The completeness and quality of the input data, along with the proper choice of algorithms and parameters, have a significant impact on prediction accuracy. Computational models frequently struggle to adequately represent the complexity of biological systems, which results in simplifications and approximations that may compromise the accuracy of the findings. The computing requirements of some in-silico techniques represent another drawback. Large biomolecular systems or lengthy simulation times can make molecular dynamics simulations computationally demanding, consuming a substantial amount of time and computing resources. For in-silico models to be dependable and predictive, validation is essential. This frequently necessitates thorough experimental validation, which can be expensive and time-consuming ¹⁰². Animal investigations are required to validate in-silico models ¹⁰³. There is still a benefit to employing animal models in spite of their drawbacks. Due to their inability to precisely represent human physiology, animal models have limitations that could result in failed drug development attempts and imprecise forecasts of in vivo adverse responses to pharmacological treatment ¹⁰⁴. It's critical to recognize the limitations of in-silico models, even though they present a potential way to speed up drug development and cut expenses.

Creating a functional and secure AI model for practical applications can be challenging. Computer simulations using personalized data are used in in-silico clinical trials to design and assess medicinal goods for regulatory approval. By simulating the effects of medical interventions in a virtual patient population using computational models, in-silico trials, also known as computer-based clinical trials, provide a revolutionary approach to medical research and development. Instead of being used in place of conventional experimental techniques, in-silico models should be viewed as supplementary instruments.

Strengths:

• Rapid screening of large compound libraries.
• Cost-effective compared to in-vitro/in-vivo
• Supports 3Rs principle in ethical research.
• Predicts multiple endpoints simultaneously.
• Highly customizable and scalable with AI models.

Limitations:

• Model accuracy is limited by training dataset quality.
• Poor extrapolation to novel chemical space.
• Limited interpretability of black-box models (e.g., deep learning).
• Lack of standardization across tools and endpoints.
• Validation and reproducibility issues limit regulatory trust.

Regulatory Acceptance and Guidelines: The in-silico toxicity regulatory environment is changing. The growing use of in-silico tools, which include a wide range of computational techniques such as databases, quantitative structure-activity relationships, pharmacophores, homology models, machine learning, and data mining, is drastically changing the pharmaceutical industry. From target identification and validation to lead optimization and ADMET (absorption, distribution, metabolism, excretion, and toxicity) property prediction, these technologies are used at many phases of drug discovery and development. By giving priority to compounds with a higher chance of success, in-silico approaches have the potential to speed up the drug development process, lower expenses, and enhance the selection of therapeutic candidates ¹⁰⁵.

Regulatory bodies throughout the world are now considering the significance of these computational tools in pharmaceutical research and have established guidelines for their proper usage and adoption as a result of the increasing dependence on them. To fully utilize the potential of in-silico screening, which poses a substantial barrier, it is crucial to remember that virtual screening pipelines can be efficiently built utilizing open-source software tools ¹⁰⁶.  A number of crucial elements must be met for in-silico tools to be approved by regulators, such as the models' validity, the methodologies' openness, and the predictions' applicability to clinical results. Standards and best practices for the use of in-silico models in drug development are being actively defined by regulatory agencies like the FDA and EMA.

Showing that the model correctly predicts the desired properties or results is a crucial part of model validation, and this calls for the utilization of high-quality experimental data for both training and testing. In addition, the models must be transparent and well-documented, with precise explanations of the assumptions, parameters, and methods employed. High testability and reproducibility of results are crucial, with a focus on methodological procedures that are clear and constraints that are thoroughly examined. The quality and applicability of the data used to construct and validate the models have a significant impact on the dependability of in-silico predictions. The application of in-silico techniques is consistent with the current trend, supported by the "3R" principle, to replace non-clinical testing with alternatives such as in-vitro or in-silico techniques.

By potentially lowering the costs and dangers of severe side effects associated with medication development, in-silico technologies can forecast drug-target interactions and drug repositioning, providing tailored therapy. The use of virtual screening to search through enormous libraries of tiny molecules for new hits with desired characteristics that can be confirmed through experimentation is one example. In-silico models can offer important insights in situations when conventional animal research is restricted because of time restrictions, ethical considerations, or inconclusive results.

When hypothesis probing is challenging because of measuring device limits or the cost of equipment, simulations of chemical systems, provided they are based on a reasonable model, can be a good substitute for experiments. Various data from animal trials and cell cultures, including high-throughput systems biology research, can be integrated into mechanistic computational models.

Finding new pharmacological targets, quantifiable biomarkers for drug action, and patient populations for which a treatment is likely to be effective or ineffective are all possible with the use of these models, which may also translate results into the context of human disease. In particular, the predictive power of in-silico approaches to detect possible medication-drug interactions is significant; it immediately leads to improved drug safety profiles and a decrease in adverse events that are seen in clinical settings ¹⁰⁷.

To encourage the prudent use of in-silico tools in the pharmaceutical sector, a number of programs and recommendations have been developed. For instance, hundreds of professionals worked with the Avicenna Support Action to create a roadmap for in-silico clinical trials, identifying use cases and obstacles to adoption. One of the most important steps in evaluating the validity of a computational model is its honest reporting. The TRIPOD (transparent reporting of a multivariable prediction model for individual prognosis or diagnosis) statement, a guideline created especially for prediction model studies, can be used to do this. These initiatives highlight the expanding understanding of the significance of defining precise regulatory frameworks for the use of in-silico techniques in contemporary drug development.

Several agencies now recognize computational models under specific conditions:

• OECD Principles for QSAR validation require defined endpoints, algorithm transparency, and applicability domain (OECD, 2007).
• REACH (EU) allows QSAR predictions if supported by robust documentation.
• US FDA accepts in-silico predictions in early submissions, especially for genotoxic impurities or carcinogenic risk assessment.
• ICH M7 encourages the use of two complementary (Q) SAR tools to evaluate mutagenic impurities in drug substances.

While these guidelines show progress, most regulatory bodies still view in-silico predictions as supportive rather than standalone evidence.

Future Directions and Challenges:

Future efforts in in-silico toxicology should focus on:

• Improved data sharing and model transparency.
• Integration with omics data (e.g., toxicogenomics).
• Development of hybrid models combining QSAR, AI, and simulation.
• Crowdsourcing and open platforms to enhance reproducibility.
• Incorporation into regulatory decision-making frameworks.

Additionally, regulatory harmonization across agencies will be essential to promote global acceptance of these tools.

CONCLUSION: A revolutionary change in drug research and safety assessment is represented by in-silico toxicity prediction methods. Making decisions more quickly, morally, and economically is made possible by their capacity to forecast harmful repercussions using chemical structure and computer modeling. Even though there are still issues with model accuracy, validation, and regulatory approval, continuous developments in artificial intelligence, data integration, and global cooperation are quickly increasing the usefulness and legitimacy of these tools. In-silico toxicology is set to become a crucial component of preclinical risk assessment and regulatory filings as the pharmaceutical industry continues to embrace digital innovation.

ACKNOWLEDGEMENT: The authors acknowledge the support of the institution of NMIMS, SPPSPTM, Mumbai in the preparation of this manuscript.

CONFLICT OF INTEREST: The authors affirm that there is no conflict of interest in this review article.

REFERENCES:

1. Schoonen WGEJ, Westerink WMA and van de Water FM: High‐Throughput Toxicity Testing in Drug Development: Aims, Strategies, and Novel Trends. In: High‐Throughput Screening Methods in Toxicity Testing Wiley 2013; 33–75; doi: 10.1002/9781118538203.ch2.
2. Segall MD and Barber C: Addressing toxicity risk when designing and selecting compounds in early drug discovery. Drug Discov Today 2014; 19(5): 688–693; doi: 10.1016/j.drudis.2014.01.006.
3. Hornberg JJ, Laursen M and Brenden N: Exploratory toxicology as an integrated part of drug discovery. Part I: Why and how. Drug Discov Today 2014; 19(8): 1131–1136; doi: 10.1016/j.drudis.2013.12.008.
4. Doke SK and Dhawale SC: Alternatives to animal testing: A review. Saudi Pharmaceutical Journal 2015; 23(3): 223–229; doi: 10.1016/j.jsps.2013.11.002.
5. Bernardo PH and Tong JC: In-silico Design of Small Molecules 2012; 25–31. doi: 10.1007/978-1-61779-349-3_3.
6. Ekins S, Mestres J and Testa B: In-silico pharmacology for drug discovery: methods for virtual ligand screening and profiling. Br J Pharmacol 2007; 152(1): 9–20; doi: 10.1038/sj.bjp.0707305.
7. Neves BJ, Braga RC and Melo-Filho CC: QSAR-Based Virtual Screening: Advances and Applications in Drug Discovery. Front Pharmacol 2018; 9. doi: 10.3389/fphar.2018.01275.
8. Speck-Planche A and Cordeiro MNDS: Multi-Target QSAR Approaches for Modeling Protein Inhibitors. Simultaneous Prediction of Activities against Biomacromolecules Present in Gram-Negative Bacteria. Curr Top Med Chem 2015; 15(18): 1801–1813. doi: 10.2174/1568026615666150506144814.
9. Rashid MBMA: Artificial intelligence effecting a paradigm shift in drug development. SLAS Technol 2021; 26(1): 3–15. doi: 10.1177/2472630320956931.
10. Lipnick RL: Outliers: their origin and use in the classification of molecular mechanisms of toxicity. Science of The Total Environment 1991; 109–110: 131–153. doi: 10.1016/0048-9697(91)90175-E.
11. Wardecki D, Dołowy M and Bober-Majnusz K: Evaluation of the usefulness of topological indices for predicting selected physicochemical properties of bioactive substances with anti-androgenic and hypouricemic activity. Molecules 2023; 28(15): 5822; doi: 10.3390/molecules28155822.
12. Guha R: On Exploring Structure–Activity Relationships. 2013; 81–94. doi: 10.1007/978-1-62703-342-8_6.
13. Oliveira PF, Guedes RC and Falcao AO: Inferring molecular inhibition potency with AlphaFold predicted structures. Sci Rep 2024; 14(1): 8252. doi: 10.1038/s41598-024-58394-z.
14. Zhu H, Zhang Y and Li W: A Comprehensive survey of prospective structure-based virtual screening for early drug discovery in the past fifteen years. Int J Mol Sci 2022; 23(24): 15961; doi: 10.3390/ijms232415961.
15. Torres PHM, Sodero ACR and Jofily P: Key topics in molecular docking for drug design. Int J Mol Sci 2019; 20(18): 4574. doi: 10.3390/ijms20184574.
16. Meng XY, Zhang HX and Mezei M: Molecular Docking: A powerful approach for structure-based drug discovery. Current Computer Aided-Drug Design 2011; 7(2): 146–157. doi: 10.2174/157340911795677602.
17. Dias R and de Azevedo W: Molecular docking algorithms. Curr Drug Targets 2008; 9(12): 1040–1047; doi: 10.2174/138945008786949432.
18. Saini M, Mehra N and Kumar G: Molecular and structure-based drug design: From Theory to Practice 2025; 121–138. doi: 10.1016/bs.apha.2025.02.004.
19. Spassov DS: Binding affinity determination in drug design: insights from lock and key, induced fit, conformational selection, and inhibitor trapping models. Int J Mol Sci 2024; 25(13): 7124. doi: 10.3390/ijms25137124.
20. NM N: Molecular docking: considerations of a low cost and suitable methodology and some successful applications. Medicinal & Analytical Chemistry International Journal 2018; 2(3). doi: 10.23880/MACIJ-16000123.
21. Pinzi L and Rastelli G: Molecular Docking: Shifting Paradigms in Drug Discovery. Int J Mol Sci 2019; 20(18): 4331. doi: 10.3390/ijms20184331.
22. Pattar SV, Adhoni SA and Kamanavalli CM: In-silico molecular docking studies and MM/GBSA analysis of coumarin-carbonodithioate hybrid derivatives divulge the anticancer potential against breast cancer. Beni Suef Univ J Basic Appl Sci 2020; 9(1): 36. doi: 10.1186/s43088-020-00059-7.
23. Permann C, Seidel T and Langer T: Greedy 3-Point Search (G3PS)—A Novel Algorithm for Pharmacophore Alignment. Molecules 2021; 26(23): 7201; doi: 10.3390/molecules26237201.
24. Vuorinen A and Schuster D: Methods for generating and applying pharmacophore models as virtual screening filters and for bioactivity profiling. Methods 2015; 71: 113–134. doi: 10.1016/j.ymeth.2014.10.013.
25. Seidel T, Wieder O and Garon A: Applications of the pharmacophore concept in natural product inspired drug design. Mol Inform 2020; 39(11). doi: 10.1002/minf.202000059.
26. Temml V and Kutil Z: Structure-based molecular modeling in SAR analysis and lead optimization. Comput Struct Biotechnol J 2021; 19: 1431–1444. doi: 10.1016/j.csbj.2021.02.018.
27. Fatima K, Khalid S and Qadeer K: In-vitro antigout potential of Alstoniascholaris flower, characterization and prospective ligand-receptor interaction of bioactive lead compound. Heliyon 2023; 9(3): 14093; doi: 10.1016/j.heliyon.2023.e14093.
28. Giordano D, Biancaniello C and Argenio MA: Drug design by pharmacophore and virtual screening approach. Pharmaceuticals 2022; 15(5): 646; doi: 10.3390/ph15050646.
29. Wieder M, Perricone U and Seidel T: Pharmacophore models derived from molecular dynamics simulations of protein-ligand complexes: a case study. Nat Prod Commun 2016; 11(10). doi: 10.1177/1934578X1601101019.
30. Temml V and Kutil Z: Structure-based molecular modeling in SAR analysis and lead optimization. Comput Struct Biotechnol J 2021; 19: 1431–1444. doi: 10.1016/j.csbj.2021.02.018.
31. Lu Z, Song G and Zhu H: DTIAM: a unified framework for predicting drug-target interactions, binding affinities and drug mechanisms. Nat Commun 2025; 16(1): 2548. doi: 10.1038/s41467-025-57828-0.
32. Abbas MKG, Rassam A and Karamshahi F: The role of ai in drug discovery. Chem Bio Chem 2024; 25(14). doi: 10.1002/cbic.202300816.
33. Chang AC: Big data in medicine: The upcoming artificial intelligence. Prog Pediatr Cardiol 2016; 43: 91–94; doi: 10.1016/j.ppedcard.2016.08.021.
34. Blanco-González A, Cabezón A and Seco-González A: The role of ai in drug discovery: challenges, opportunities, and strategies. Pharmaceuticals 2023; 16(6): 891. doi: 10.3390/ph16060891.
35. Su J, Xin C and Shang A: Artificial intelligence in drug discovery: A comprehensive review with a case study on hyperuricemia, gout arthritis, and hyperuricemic nephropathy 2025.
36. Ekins S, Mestres J and Testa B: In-silico pharmacology for drug discovery: methods for virtual ligand screening and profiling. Br J Pharmacol 2007; 152(1): 9–20; doi: 10.1038/sj.bjp.0707305.
37. Bajorath J: Computer-aided drug discovery. F1000Res 2015; 4: 630. doi: 10.12688/f1000research.6653.1.
38. Serrano DR, Luciano FC and Anaya BJ: Artificial Intelligence (AI) applications in drug discovery and drug delivery: revolutionizing personalized medicine. Pharmaceutics 2024; 16(10): 1328; doi: 10.3390/pharmaceutics16101328.
39. Kim H, Kim E and Lee I: Artificial intelligence in drug discovery: a comprehensive review of data-driven and machine learning approaches. Biotechnology and Bioprocess Engineering 2020; 25(6): 895–930.
40. Ferreira FJN and Carneiro AS: AI-Driven drug discovery: a comprehensive review. ACS Omega 2025; 10(23): 23889–23903. doi: 10.1021/acsomega.5c00549.
41. Romm EL and Tsigelny IF: Artificial Intelligence in Drug Treatment. Annu Rev Pharmacol Toxicol 2020; 60(1): 353–369. doi: 10.1146/annurev-pharmtox-010919-023746.
42. Kant S, Deepika and Roy S: Artificial intelligence in drug discovery and development: transforming challenges into opportunities. Discover Pharmaceutical Sciences 2025; 1(1): 7. doi: 10.1007/s44395-025-00007-3.
43. Rehman AU, Li M and Wu B: Role of artificial intelligence in revolutionizing drug discovery. Fundamental Research 2025; 5(3): 1273–1287. doi: 10.1016/j.fmre.2024.04.021.
44. Qureshi R, Irfan M and Gondal TM: AI in drug discovery and its clinical relevance. Heliyon 2023; 9(7): 17575. doi: 10.1016/j.heliyon.2023.e17575.
45. Taherdoost H and Ghofrani A: AI’s role in revolutionizing personalized medicine by reshaping pharmacogenomics and drug therapy. Intelligent Pharmacy 2024; 2(5): 643–650. doi: 10.1016/j.ipha.2024.08.005.
46. Niazi SK and Mariam Z: Artificial intelligence in drug development: reshaping the therapeutic landscape. Ther Adv Drug Saf 2025; 16. doi: 10.1177/20420986251321704.
47. Bédard P, Gauvin S and Ferland K: Innovative human three-dimensional tissue-engineered models as an alternative to animal testing. Bioengineering 2020; 7(3): 115. doi: 10.3390/bioengineering7030115.
48. Koch U, Hamacher M and Nussbaumer P: Cheminformatics at the interface of medicinal chemistry and proteomics. Biochimica et Biophysica Acta (BBA) - Proteins and Proteomics 2014; 1844(1): 156–161. doi: 10.1016/j.bbapap.2013.05.010.
49. Ancuceanu R, Hovanet MV and Anghel AI: Computational models using multiple machine learning algorithms for predicting drug hepatotoxicity with the dilirank dataset. Int J Mol Sci 2020; 21(6): 2114; doi: 10.3390/ijms21062114.
50. Kusko R and Hong H: Computational Toxicology Promotes Regulatory Science 2019; 1–11; doi: 10.1007/978-3-030-16443-0_1.
51. Petrou A, Geronikaki A and Kartsev V: N-Derivatives of (Z)-Methyl 3-(4-Oxo-2-thioxothiazolidin-5-ylidene) methyl)-1H-indole-2-carboxylates as Antimicrobial agents—in-silico and in-vitro evaluation. Pharmaceuticals 2023; 16(1): 131. doi: 10.3390/ph16010131.
52. Tran TT Van, Surya Wibowo A and Tayara H: Artificial intelligence in drug toxicity prediction: recent advances, challenges, and future perspectives. J Chem Inf Model 2023; 63(9): 2628–2643. doi: 10.1021/acs.jcim.3c00200.
53. Hao Y, Romano JD and Moore JH: Knowledge-guided deep learning models of drug toxicity improve interpretation. Patterns 2022; 3(9): 100565. doi: 10.1016/j.patter.2022.100565.
54. Masarone S, Beckwith KV and Wilkinson MR: Advancing predictive toxicology: overcoming hurdles and shaping the future. Digital Discovery 2025; 4(2): 303–315. doi: 10.1039/D4DD00257A.
55. Guha R: On Exploring Structure–Activity Relationships 2013; 81–94. doi: 10.1007/978-1-62703-342-8_6.
56. Petrou A, Geronikaki A and Kartsev V: N-Derivatives of (Z)-Methyl 3-(4-Oxo-2-thioxothiazolidin-5-ylidene) methyl)-1H-indole-2-carboxylates as Antimicrobial agents in-silico and in-vitro evaluation. Pharmaceuticals 2023; 16(1): 131. doi: 10.3390/ph16010131.
57. Tran TT Van, Surya Wibowo A and Tayara H: Artificial intelligence in drug toxicity prediction: recent advances, challenges, and future perspectives. J Chem Inf Model 2023; 63(9): 2628–2643; doi: 10.1021/acs.jcim.3c00200.
58. Bédard P, Gauvin S and Ferland K: Innovative human three-dimensional tissue-engineered models as an alternative to animal testing. Bioengineering 2020; 7(3): 115. doi: 10.3390/bioengineering7030115.
59. Vo AH, Van Vleet TR and Gupta RR: An overview of machine learning and big data for drug toxicity evaluation. Chem Res Toxicol 2020; 33(1): 20–37; doi: 10.1021/acs.chemrestox.9b00227.
60. Masarone S, Beckwith KV and Wilkinson MR: Advancing predictive toxicology: overcoming hurdles and shaping the future. Digital Discovery 2025; 4(2): 303–315; doi: 10.1039/D4DD00257A.
61. Abdel-Aziz AA-M, El-Azab AS and Alanazi AM: Synthesis and potential antitumor activity of 7-(4-substituted piperazin-1-yl)-4-oxoquinolines based on ciprofloxacin and norfloxacin scaffolds: in-silico studies. J Enzyme Inhib Med Chem 2016; 31(5): 796–809. doi: 10.3109/14756366.2015.1069288.
62. Xiong G, Wu Z and Yi J: ADMETlab 2.0: an integrated online platform for accurate and comprehensive predictions of ADMET properties. Nucleic Acids Res 2021; 49(1): 5–14. doi: 10.1093/nar/gkab255.
63. Peón A, Dang CC and Ballester PJ: How Reliable Are Ligand-Centric Methods for Target Fishing? Front Chem 2016; 4. doi: 10.3389/fchem.2016.00015.
64. Jang IJ: Artificial intelligence in drug development: clinical pharmacologist perspective. Transl Clin Pharmacol 2019; 27(3): 87. doi: 10.12793/tcp.2019.27.3.87.
65. Sakamuru S, Huang R and Xia M: Use of Tox21 Screening Data to Evaluate the COVID-19 Drug Candidates for Their Potential Toxic Effects and Related Pathways. Front Pharmacol 2022; 13. doi: 10.3389/fphar.2022.935399.
66. Abdolmaleki A, Ghasemi J and Ghasemi F: Computer aided drug design for multi-target drug design: SAR /QSAR, molecular docking and pharmacophore methods. Curr Drug Targets 2017; 18(5): 556–575; doi: 10.2174/1389450117666160101120822.
67. Bai C, Wu L and Li R: Machine learning‐enabled drug‐induced toxicity prediction. Advanced Science 2025; 12(16). doi: 10.1002/advs.202413405.
68. Siramshetty V, Williams J and Nguyễn ÐT: Validating ADME QSAR Models Using Marketed Drugs. SLAS Discovery 2021; 26(10): 1326–1336. doi: 10.1177/24725552211017520.
69. Wardecki D, Dołowy M and Bober-Majnusz K: Evaluation of the usefulness of topological indices for predicting selected physicochemical properties of bioactive substances with anti-androgenic and hypouricemic activity. Molecules 2023; 28(15): 5822; doi: 10.3390/molecules28155822.
70. Balani S, Miwa G and Gan LS: Strategy of utilizing in-vitro and in-vivo adme tools for lead optimization and drug candidate selection. Curr Top Med Chem 2005; 5(11): 1033–1038. doi: 10.2174/156802605774297038.
71. Lai Y, Chu X and Di L: Recent advances in the translation of drug metabolism and pharmacokinetics science for drug discovery and development. Acta Pharm Sin B 2022; 12(6): 2751–2777. doi: 10.1016/j.apsb.2022.03.009.
72. Komura H, Watanabe R and Mizuguchi K: The trends and future prospective of in-silico models from the viewpoint of adme evaluation in drug discovery. Pharmaceutics 2023; 15(11): 2619. doi: 10.3390/pharmaceutics15112619.
73. Wardecki D, Dołowy M and Bober-Majnusz K: Evaluation of the usefulness of topological indices for predicting selected physicochemical properties of bioactive substances with anti-androgenic and hypouricemic activity. Molecules 2023; 28(15): 5822; doi: 10.3390/molecules28155822.
74. Zhang Z and Tang W: Drug metabolism in drug discovery and development. Acta Pharm Sin B 2018; 8(5): 721–732. doi: 10.1016/j.apsb.2018.04.003.
75. Miller D, Fraczkiewicz R and Woltosz W: Novel ADMET design tool for chemists. J Cheminform 2011; 3(1): 9. doi: 10.1186/1758-2946-3-S1-P9.
76. Guan L, Yang H and Cai Y: ADMET-score – a comprehensive scoring function for evaluation of chemical drug-likeness. Medchemcomm 2019; 10(1): 148–157. doi: 10.1039/C8MD00472B.
77. Xiong G, Wu Z and Yi J: ADMETlab 2.0: an integrated online platform for accurate and comprehensive predictions of ADMET properties. Nucleic Acids Res 2021; 49(1): 5–14; doi: 10.1093/nar/gkab255.
78. Chebieb A, Ziani-Cherif C and Bellifa K: In-silico Studies toward the Improvement of the Antibacterial Activity of Pristinamycin IIB. In: The 25th International Electronic Conference on Synthetic Organic Chemistry MDPI: Basel Switzerland 2021; 13. doi: 10.3390/ecsoc-25-11703.
79. Siddique F, Anwaar A and Bashir M: Revisiting methotrexate and phototrexate Zinc15 library-based derivatives using deep learning in-silico drug design approach. Front Chem 2024; 12. doi: 10.3389/fchem.2024.1380266.
80. Wardecki D, Dołowy M and Bober-Majnusz K: Evaluation of the usefulness of topological indices for predicting selected physicochemical properties of bioactive substances with anti-androgenic and hypouricemic activity. Molecules 2023; 28(15): 5822; doi: 10.3390/molecules28155822.
81. Miller D, Fraczkiewicz R and Woltosz W: Novel ADMET design tool for chemists. J Cheminform 2011; 3(1): 9. doi: 10.1186/1758-2946-3-S1-P9.
82. Guan L, Yang H and Cai Y: ADMET-score – a comprehensive scoring function for evaluation of chemical drug-likeness. Medchemcomm 2019; 10(1): 148–157. doi: 10.1039/C8MD00472B.
83. Vora LK, Gholap AD and Jetha K: Artificial Intelligence in Pharmaceutical Technology and Drug Delivery Design. Pharmaceutics 2023; 15(7): 1916. doi: 10.3390/pharmaceutics15071916.
84. Chen H, Engkvist O and Kogej T: Compound Properties and Their Influence on Drug Quality. In: The Practice of Medicinal Chemistry Elsevier 2015; 379–393. doi: 10.1016/B978-0-12-417205-0.00015-8.
85. Jang IJ: Artificial intelligence in drug development: clinical pharmacologist perspective. Transl Clin Pharmacol 2019; 27(3): 87. doi: 10.12793/tcp.2019.27.3.87.
86. Neves BJ, Braga RC and Melo-Filho CC: QSAR-Based virtual screening: advances and applications in drug discovery. Front Pharmacol 2018; 9. doi: 10.3389/fphar.2018.01275.
87. Tuccinardi T: Docking-based virtual screening: recent developments. Comb Chem High Throughput Screen 2009; 12(3): 303–314; doi: 10.2174/138620709787581666.
88. Ekins S, Mestres J and Testa B: In-silico pharmacology for drug discovery: methods for virtual ligand screening and profiling. Br J Pharmacol 2007; 152(1): 9–20. doi: 10.1038/sj.bjp.0707305.
89. Giordano D, Biancaniello C and Argenio MA: Drug Design by Pharmacophore and Virtual Screening Approach. Pharmaceuticals 2022; 15(5): 646. doi: 10.3390/ph15050646.
90. Rashid MBMA: Artificial intelligence effecting a paradigm shift in drug development. SLAS Technol 2021; 26(1): 3–15. doi: 10.1177/2472630320956931.
91. Neves BJ, Braga RC and Melo-Filho CC: QSAR-Based virtual screening: advances and applications in drug discovery. Front Pharmacol 2018; 9. doi: 10.3389/fphar.2018.01275.
92. Suay-García B, Bueso-Bordils JI and Falcó A: Virtual Combinatorial chemistry and pharmacological screening: a short guide to drug design. Int J Mol Sci 2022; 23(3): 1620. doi: 10.3390/ijms23031620.
93. Zhu T, Cao S and Su PC: Hit identification and optimization in virtual screening: practical recommendations based on a critical literature analysis. J Med Chem 2013; 56(17): 6560–6572. doi: 10.1021/jm301916b.
94. Crofton KM, Bassan A and Behl M: Current status and future directions for a neurotoxicity hazard assessment framework that integrates in-silico approaches. Computational Toxicology 2022; 22: 100223. doi: 10.1016/j.comtox.2022.100223.
95. Siepmann J and Siepmann F: Modeling of diffusion controlled drug delivery. Journal of Controlled Release 2012; 161(2): 351–362; doi: 10.1016/j.jconrel.2011.10.006.
96. Pattar SV, Adhoni SA and Kamanavalli CM: In-silico molecular docking studies and MM/GBSA analysis of coumarin-carbonodithioate hybrid derivatives divulge the anticancer potential against breast cancer. Beni Suef Univ J Basic Appl Sci 2020; 9(1): 36. doi: 10.1186/s43088-020-00059-7.
97. Bajorath J: Computer-aided drug discovery. F1000Res 2015; 4: 630. doi: 10.12688/f1000research.6653.1.
98. Bédard P, Gauvin S and Ferland K: Innovative human three-dimensional tissue-engineered models as an alternative to animal testing. Bioengineering 2020; 7(3): 115. doi: 10.3390/bioengineering7030115.
99. Viceconti M, Henney A and Morley-Fletcher E: In-silico clinical trials: how computer simulation will transform the biomedical industry. Int J Clin Trials 2016; 3(2): 37. doi: 10.18203/2349-3259.ijct20161408.
100. Filipić B, Ušjak D and Rambaher MH: Evaluation of novel compounds as anti-bacterial or anti-virulence agents. Front Cell Infect Microbiol 2024; 14. doi: 10.3389/fcimb.2024.1370062.
101. Bhargava H, Sharma A and Suravajhala P: Chemogenomic approaches for revealing drug target interactions in drug discovery. Curr Genomics 2021; 22(5): 328–338; doi: 10.2174/1389202922666210920125800.
102. Ponnarengan H, Rajendran S and Khalkar V: Data-Driven healthcare: the role of computational methods in medical innovation. Computer Modeling in Engineering & Sciences 2025; 142(1): 1–48.
103. Sharma K, Arora T and Joshi V: Substitute of animals in drug research: An approach towards fulfillment of 4R′s. Indian J Pharm Sci 2011; 73(1): 1. doi: 10.4103/0250-474X.89750.
104. Zheng F, Xiao Y and Liu H: Patient‐Specific Organoid and Organ‐on‐a‐Chip: 3D Cell‐Culture Meets 3D Printing and Numerical Simulation. Adv Biol 2021; 5(6). doi: 10.1002/adbi.202000024.
105. Giordano D, Biancaniello C and Argenio MA: Drug Design by Pharmacophore and Virtual Screening Approach. Pharmaceuticals 2022; 15(5): 646. doi: 10.3390/ph15050646.
106. Glaab E: Building a virtual ligand screening pipeline using free software: a survey. Brief Bioinform 2016; 17(2): 352–366. doi: 10.1093/bib/bbv037.
107. Filipić B, Ušjak D and Rambaher MH: Evaluation of novel compounds as anti-bacterial or anti-virulence agents. Front Cell Infect Microbiol 2024; 14. doi: 10.3389/fcimb.2024.1370062.
     

     ---

     How to cite this article:

     Shahi S and Sharma S: In-silico toxicity prediction tools: a review of techniques and applications. Int J Pharm Sci & Res 2026; 17(4): 1058-72. doi: 10.13040/IJPSR.0975-8232.17(4).1058-72.

     ---

     All © 2026 are reserved by International Journal of Pharmaceutical Sciences and Research. This Journal licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License.