Добавил:
kiopkiopkiop18@yandex.ru t.me/Prokururor I Вовсе не секретарь, но почту проверяю Опубликованный материал нарушает ваши авторские права? Сообщите нам.
Вуз: Предмет: Файл:
Ординатура / Хирургия / Библиотека им академика М.И. Перельмана / Книга_5440_Библиотеки_им_академика_М_И_Перельмана.pdf
Скачиваний:
0
Добавлен:
10.10.2026
Размер:
9 Мб
Скачать
☆
293
The ongoing exploration in the pursuit of new bioactive substances with industrial applications
remains a focal point of scientific research, driven by the inherent pharmacophoric structures, pharma-
cokinetic attributes, and distinct chemical space exhibited by these compounds. The systematic investi-
gation of natural sources for the extraction of valuable molecules, essential for the development of
commercially and industrially relevant products, poses a substantial challenge in the field of bioprospect-
ing. Notably, the advent of advanced VS strategies has revolutionized the discovery process by employing
in silico analyses of expansive compound libraries. This tactical approach not only makes it easier to
conduct a thorough analysis of the chemical-based space, pharmaceutical dynamics, and pharmacoki-
netic characteristics of these molecules but it also greatly reduces the time, infrastructure, and financial
costs associated with the complex process of locating and describing novel chemical entities [48].
The principal uses for online chemical screening involve the careful selection from chemical
libraries of a limited but receptor-relevant fraction with an emphasis on maximum chemical diver-
sity. To achieve a balance between accuracy and computing efficiency throughout the hit enrich-
ment process, our earlier research demonstrated the effectiveness of integrating ligand-centric and
receptor-centric VS approaches. In this work, we present a “progressive distributed docking”
methodology designed to improve the VS procedure by iteratively combining shape-matching and
docking phases. First, 3D templates were used for shape comparisons within the chemical library.
These templates were derived from known ligands with poor docking scores. Then, molecules with
low receptor docking scores and good template shape matches were repeatedly chosen for addi-
tional rounds of shape searching and docking. This VS procedure is iterative and was verified
through the enrichment of compounds from a carefully selected subset of chemical libraries that
were pertinent to phosphoinositide 3-kinase and peroxisome proliferator-activated receptors. The
outcomes showed how this progressive distributed docking technique improves lead-hopping pro-
cedures by increasing the chemical variety of the chosen virtual hit cohort [49].
13.6.2 QSAR (Quantitative Structure–Activity Relationship) Models for
Optimizing Biological Properties
The foundation of computational science is the QSAR, which establishes relationships between
the chemical structures and biological activities of various molecules. QSAR models are essential
for predicting the biological activity of organic compounds against particular targets, which helps
identify possible candidates for drugs. These models are useful for the effective VS of natural com-
pounds for various pharmacological profiles, especially in the field of antiparasitic activity. Notably,
nevertheless, the use of QSAR models to forecast the activity against several parasite species using
a single model is still a major and continuous problem, highlighting the necessity for further devel-
opments in this computational approach [50].
Drug development activities greatly benefit from the use of QSAR models, which are crucial
instruments for forecasting the biological effects of compounds. These models are useful for a wide
range of pharmacological actions and toxicities, which helps find effective medications more
quickly. QSAR models play a key role in the setting of natural product drug development by mak-
ing it easier to anticipate the biological effects of phytochemicals, or molecules originating from
plants [51]. With QSAR models’ ability to simultaneously predict pharmacological activity,
toxicities, and safety profiles based on the integration of varied chemical and biological data, their
application becomes especially significant in the design of improved pharmaceuticals derived
from phytochemicals. Consequently, by using their predictive power to identify the biological
effects of phytochemicals, QSAR models become indispensable components of the seamless
13.6 In Silico         
        294
integration of natural product drug development to help simplify the creation of powerful yet
secure medications [52].

13.6.3 High-Throughput Screening Methods for Efficient Compound Selection

The seamless integration of solid-phase extraction, high-throughput parallel preparative high-
performance liquid chromatography, and automated flash chromatography highlights the cru-
cial importance of HTS approaches. All of these approaches result in libraries that are composed
of one to five compounds per well, roughly. To support this, high-throughput parallel liquid
chromatography-MS-evaporative light scattering detection systems are used to do thorough
library analysis before biological screening [53]. Target-based workflows in natural product
drug discovery projects are significantly more efficient thanks to this novel method, which
quickly identifies natural product ligands that target human drug receptors without the need
for fractionation [54].

13.6.4 Molecular Dynamics Simulations for Predicting Solubility and Stability

One essential component that is considered essential for all ADME research is the evaluation of
water solubility. Acknowledged as a crucial factor, water solubility defines how a medicine is
absorbed orally and how bioavailable it is. Since a significant fraction of pharmaceuticals now on
the market (about 40%) and a notable majority of compounds in the development stage (around
75%) have low solubility in water, it is imperative to identify this feature as soon as possible. This
proactive approach illustrates the critical role that aqueous solubility plays in determining the suc-
cess trajectory of drug candidates. It not only informs strategic decisions but also has the potential
to minimize failures in the pharmaceutical development process [55].
A multitude of experimental techniques, such as modifications of the shake-flask method and
the CheqSol methodology, have been utilized to ascertain the solubility of various substances in
water. However, the experimental determination turns out to be difficult, expensive, and time-
consuming, especially when dealing with the large chemical libraries that are utilized in HTS. As
a result, early in the drug discovery and development process, in silico prediction of water solubil-
ity by quantitative structure–property relationship (QSPR) has gained significance. Although sev-
eral QSPR models have been created over the years, their effectiveness on various solubility
datasets has exposed some of the shortcomings of certain prediction techniques [55]. Notably, the
inconsistent data sources are the primary cause of these restrictions, with an average root-mean-
square error (RMSE) of 0.6–0.7 log S units. According to recent research, the insufficiency in there-
fore, more accurate approaches, such as advanced machine learning (ML) algorithms like random
forests (RF), support vector machines (SVM), k-nearest neighbors (k-NN), convolutional, and
recurrent networks, are recommended. Solubility prediction accuracy is not only related to the
quality of experimental data. These ML techniques have been shown to perform on par with or
better than conventional techniques in terms of aqueous solubility prediction [56].
Using sophisticated prediction models, especially in the field of ML, is a valuable tool for predict-
ing a compound’s solubility, which is a critical factor in the early stages of drug discovery and
development. Predicting the complex characteristics of a compound’s fate in a biological system is
made even more accurate by adding molecular descriptors and physicochemical parameters to
solubility predictions. Its usefulness is expanded by this comprehensive method to include critical
parameter prediction, including ADMET [57]. Drug development initiatives can make well-
informed decisions at an early stage and identify promising candidates more quickly by
295
incorporating these predictive tools. This also makes it easier to comprehend a compound’s
potential within the complex field of pharmacological and toxicological research [58].

13.6.5 ADMET Attributes Predicted In Silico

Computational models are used in silico ADME-Tox prediction to evaluate a drug candidate’s ADMET
characteristics. Using these models is essential for strategically ranking possible compounds, which
lowers development costs for new drugs and lowers attrition rates. However, the intricate nature of
physiological processes poses challenges in the generation of reliable and accurate prediction mod-
els [59]. The KnowItAll system offers a comprehensive approach, providing both real-number and
categorical classification predictions, thereby addressing concerns about the reliability of predic-
tions. While in silico models prove invaluable for early-stage screening, it is imperative to acknowl-
edge that they do not serve as outright replacements for in vivo or in vitro methods, emphasizing the
necessity of a multifaceted approach in comprehensively assessing a drug candidate’s viability [60].
13.6.6 Computational Tools for Predicting Pharmacokinetics and
Pharmacodynamics (PK/PD)
A predominant challenge in drug development lies in the substantial rate of failures attributed to
issues with (ADME/Tox) properties of candidate compounds, accounting for over half of the set-
backs. The strategic integration of in silico tools to predict ADME/Tox and physicochemical prop-
erties emerges as a promising avenue to mitigate attrition rates in pharmaceutical research and
development. This technology, exemplified by the KnowItAll computational environment devel-
oped by Bio-Rad Laboratories, Inc., has the potential to streamline the pharmaceutical R&D pipe-
line by effectively prioritizing candidate compounds. A noteworthy aspect of KnowItAll is its
commitment to addressing concerns about the reliability of property predictions. Within this plat-
form, various ADME/Tox predictors are encoded, providing the capability to verify these predic-
tions using internal data and models, if any. Moreover, the system allows for the construction of a
“consensus” model, surpassing individual predictive models in efficacy. This sophisticated envi-
ronment adeptly handles both real-number and categorical classification predictions, contributing
to a comprehensive and reliable approach to navigating the intricacies of drug development [61].
NPs and their derivatives are important sources for the development of new drugs; however, cur-
rent in silico target prediction techniques frequently cannot distinguish between NPs and artificial
compounds. Because NPs differ from their synthetic counterparts inherently, it is necessary to
develop target prediction models that are specific to NPs. To close this gap, a study was conducted
where four unique datasets were created, namely, the NP dataset, NPs and their first-class deriva-
tives dataset, NPs and all derivatives dataset, and the ChEMBL26 compounds dataset – by curating
activity data from open databases and covering a range of NP scenarios [62]. To optimize perfor-
mance, systematically investigated conditions like activity thresholds and input features using
eight ML techniques for NP target prediction. The most appropriate dataset for developing reliable
NP-specific models turned out to be the one that included all of the derivatives of NPs. Consensus
models, specifically the feedforward neural network (FNN) and SVM consensus model, and voting
models were also applied, and this improved prediction performance considerably. Extensive
assessments on external validation sets confirmed that these models performed better than con-
ventional models that were trained on all of the ChEMBL26 compounds. The effectiveness of NP-
specific target prediction models in boosting drug discovery efforts is highlighted by this
all-encompassing strategy [62].
13.6 In Silico         
        296
Predicting cytotoxicity is essential in drug development to mitigate late-stage toxicity issues.
In silico methods, particularly deep learning, are crucial for early design processes, reducing time,
costs, and reliance on animal testing. Leveraging a comprehensive dataset, a neural network model
achieves a balanced accuracy exceeding 70%, demonstrating efficacy comparable to RF. While neu-
ral networks lack interpretability, the deep Taylor decomposition method is explored to identify
toxicophores responsible for cytotoxic effects [63].
The escalating release of thousands of anthropogenic chemicals into the environment necessi-
tates robust methods for rapidly screening and predicting their potential toxicity, thereby mitigat-
ing adverse impacts on human and environmental health. Computational approaches, particularly
molecular docking, are evaluated here for screening the toxicity of diverse xenobiotic substances,
which include chemicals produced by the chemical industry, pollution, medications, and
insecticides. The methodology involves predicting binding energy between pollutants and care-
fully selected receptors, assuming toxicity correlates with interference in biochemical pathways.
One of the method’s main advantages is its speed at producing interaction maps, which provide
molecular-level information on possible perturbation pathways and help choose chemicals for
additional testing. When contrasted as scoring functions, Autodock Vina and the ML scoring func-
tion RF-Score-VS exhibit encouraging results but with limitations related to scoring function
accuracy [64].
The VirtualToxLab is an in silico tool that evaluates the carcinogenicity, cardiotoxicity, and endo-
crine and metabolic disruption potentials of medications, chemicals, and natural items. Using an
automated approach, the system provides real-time 3D/4D mechanistic interpretation in accord-
ance with the Setubal principles for computational toxicology by simulating and quantifying the
binding of small compounds to a set of 16 proteins implicated in inducing deleterious effects.
Interestingly, the “ab initio” protocol operates in client–server mode, is globally applicable, and is
free for use by academic and nonprofit institutions. It also does not require training data. Molecular
dynamics simulations are made possible by the technology’s thermodynamic estimate of binding
affinity, which enables the investigation of the kinetic stability of ligand–protein complexes [65].
With an emphasis on Trypanosoma brucei, the neglected tropical disease that causes human
African trypanosomiasis, the VirtualToxLab plays a key role in investigating novel medication can-
didates derived from natural products and utilizing their chemical diversity to address issues with
current chemotherapeutic drugs [66].
Natural products have garnered increasing attention in drug discovery for their potential thera-
peutic advantages. The integration of ADME/Tox studies is imperative to evaluate the safety and
efficacy of these compounds, given their structural diversity and distinctive biological activities.
While direct evidence on the amalgamation of ADME/Tox remains limited, ensuring the safety
and effectiveness of these compounds is paramount before advancing them into pharmaceutical
drugs. This strategic incorporation of ADME/Tox studies contributes to the comprehensive under-
standing and optimization of natural products for therapeutic development [67].
For a very long time, natural products made from plants, fungi, and bacteria have been a valuable
source of bioactive substances. Notably, small molecule natural products or their chemically modi-
fied analogs of FDA-approved medications; the remaining are synthetic drugs that target the same
biomolecules as natural products. Despite improvements in bioinformatics analysis and HTS stream-
lining certain aspects of the drug discovery pipeline, the identification of specific functions, such as
antibiotic activity, remains a bottleneck, requiring the laborious processes of production, purifica-
tion, and assaying. The traditional sequence of assaying, production, and purification introduces the
challenge of rediscovering known molecules, hindering the efficiency of natural product utilization.
Although MS- and NMR-based methods offer some enhancements, they do not eliminate the
297
fundamental requirement of having molecules for testing. Thus, the journey from the activity in a
bacterial culture extract to the discovery of a novel active molecule remains a time-consuming
impediment in harnessing natural products for antibiotic and therapeutic agent discovery [68].
A crucial phase in the drug development process is to effectively identify prospective biosyn-
thetic gene clusters (BGCs) among the expanding array of bacterial genomes. However, the exist-
ing genome mining tools mostly concentrate on comparing BGCs to known natural products. This
means that the present technique depends on gene-to-structure links. The problem of finding the
molecules with useful functionalities has not been sufficiently addressed by prioritizing based on
structural novelty, despite the fact that these techniques have identified over 147,000 BGC
sequences. Many intriguing compounds from an architectural perspective have unknown func-
tions. The potential to forecast a natural product’s activity based on its BGC creates opportunities
to focus search efforts on those most likely to produce products with the desired activities, thus
changing the nature of natural product research [68].
13.6.7 In Silico Prediction of Biosynthetic Pathways and Identification of Potential
Bioactive Compounds
Chemical reaction predictions without explicit rules can now be made, thanks to recent develop-
ments in deep learning techniques. Sequential models use string representations of chemicals,
such as SMILES. Especially in retrosynthesis prediction tasks, these rule-free models have outper-
formed their rule-based counterparts in terms of performance and potential for generalization.
Retrosynthetic pathway planning through methods has been made possible by single-step ret-
rosynthesis prediction; nevertheless, the effectiveness is limited by the requirement for expensive
online resource estimates. The Retro method is a recent development that combines an AND–OR
tree-based searching strategy with deep learning instruction to demonstrate improved planning
efficiency and solution quality. On the other hand, using multistep planning algorithms for natural
product retrosynthetic planning presents particular difficulties since pathways for biosynthesis
have more stages, more branching ratios, and less data available [69].
In the pursuit of understanding biosynthetic pathways NPs, where comprehensive information
is often elusive, the development of a user-friendly toolkit, BioNavi-NP, proves invaluable. This
toolbox uses both general organic and biosynthetic reactions to develop a single-step bioretrosyn-
thesis prediction model using an end-to-end transformer neural network. Utilizing a planning
algorithm effectively explores likely biosynthetic pathways for both NPs and chemicals that resem-
ble them. Comprehensive tests show that BioNavi-NP is 1.7 times more accurate than current rule-
based methods in identifying pathways of biosynthesis chemicals. Remarkably, the toolset
effectively pinpoints biologically likely routes for intricate nanoparticles derived from current
research. With the help of openly available curated datasets, trained models, and BioNavi-NP, bio-
synthetic pathway reconstruction for NPs can be made easier [69].
NPs stand as pivotal entities in drug discovery, contributing to over 50% of FDA-approved drugs
and demonstrating distinct selectivity toward cellular targets. The unique properties of biologically
active natural products make them promising candidates for influencing disease-related pathways
and reshaping biological networks from a diseased to a healthy state. While large-scale network
analyses, including drug–target networks, protein–protein interaction networks, metabolic net-
works, and disease pathways, have unveiled the action mechanisms of bioactive compounds, exist-
ing studies often focus on a limited number of molecules. With vast chemical diversity, NPs offer
unparalleled potential for discovering diverse bioactive molecules. Despite sporadic analyses on
aspects like chemical diversity, property distribution, molecular scaffold, and chemical space,
13.6 In Silico         
        298
comprehensive statistics comparing NPs with other compound types have been scarce due to the
challenges of obtaining extensive datasets encompassing structures and annotations [70].
A myriad of antibiotics and therapeutic drugs, including penicillin, erythromycin, tetracycline,
tacrolimus, cyclosporine (an immunosuppressant), artemisinin (an antimalarial), and acarbose
(a diabetes medication), belong to the class of natural products. Originating from microorganisms
or plants, these compounds are often categorized as “secondary metabolites” or “specialized
metabolites” due to their biosynthetic pathways being distinct from essential growth and
reproductive processes. BGCs are collections of genes essential for the synthesis of these substances
in microbes and fungi. These BGCs contain all of the genetic repertoire needed for precursors in
the process of biosynthesis, scaffold assembly, modification (tailoring), resistance, export, and
regulation. This group makes it easier to identify entire pathways by showing how a single gene
contributes to biosynthesis. On the other hand, in plants, the genes involved in specific pathways’
production are dispersed throughout the genome, requiring other experimental information, like
co-expression analysis, to identify [71].
Beginning in the early 2000s, the use of predictions in natural product biosynthesis was made
possible by the introduction of computational tools, such as the search engine. Public tools that
followed were made possible by Ecopia’s 2003 introduction of its proprietary search engine and
database. Most notably, SEARCHPKS automates the process of identifying polyketide synthases’
(PKSs) catalytic domains. The advent of antiSMASH, an open-source genome mining platform
that combined and enhanced the features of earlier tools, in 2011 marked a crucial turning point.
In addition to providing an intuitive web interface, antiSMASH made larger-scale genome mining
studies feasible for researchers lacking in-depth knowledge of computational biology. Considering
antiSMASH has steadily grown, offering a wide range of tools and databases for comparative
genomics and automated genome mining across multiple classes of secondary metabolites. The
antiSMASH pipeline can be used for both fungal and bacterial genome analysis; a special branch
called “fungiSMASH” is dedicated to fungus analysis. Moreover, plantiSMASH, a variation of ant-
iSMASH, adds support for co-expression analysis and plant-specific features such as modified
Hidden Markov Model profiles and cluster identification logic.
Computational approaches have become increasingly indispensable, presenting cost-effective
and efficient means to identify and optimize potential drug candidates. These approaches encom-
pass diverse methodologies such as molecular simulations, ML, and predictive modeling, collec-
tively serving to streamline the intricate drug discovery process and curtail associated costs.
Notwithstanding the inherent challenges, computational methods hold immense promise by
effectively narrowing down the spectrum of compounds under consideration. These tools prove
instrumental in supporting decision-makers, facilitating multiparameter optimization, and thereby
guiding the judicious selection and design of compounds with optimal prospects for success [72].

13.7 Formulation Challenges with Natural Products

Phytochemicals, bioactive compounds derived from plants, have emerged as promising candidates
in drug discovery due to their diverse pharmacological activities [73]. However, their translation
from plant sources to effective drugs faces several pharmacokinetic limitations, impeding their
progress in drug development [74].
One prominent challenge is the poor bioavailability of phytochemicals, a crucial factor influenc-
ing their efficacy. Bioavailability refers to the proportion of the administered dose that reaches the
      299
systemic circulation in an active form. Many phytochemicals exhibit low oral bioavailability owing
to factors such as poor solubility, extensive metabolism, and limited absorption [75]. The hydro-
phobic nature of certain phytochemicals hampers their dissolution in the aqueous environment of
the gastrointestinal tract, leading to reduced absorption and systemic availability [76]. In addition,
first-pass metabolism in the liver further diminishes the bioavailability of phytochemicals, result-
ing in suboptimal therapeutic outcomes [75].
The distribution of phytochemicals within the body is also a significant hurdle in drug discov-
ery [77]. The lipophilicity or hydrophilicity of these compounds influences their ability to traverse
biological membranes and reach target tissues. Lipophilic phytochemicals may accumulate in adi-
pose tissue, reducing their bioavailability to other organs. Conversely, hydrophilic compounds
may face challenges in crossing cell membranes, limiting their distribution to specific cellular
compartments [78]. Moreover, the selective permeability of physiological barriers, such as the
blood–brain barrier, presents an additional obstacle for phytochemicals targeting the central nerv-
ous system [79].
Metabolism is a critical aspect of pharmacokinetics that profoundly affects the fate of phy-
tochemicals in the body. Cytochrome P
450
enzymes in the liver play a pivotal role in the bio-
transformation of many phytochemicals [80]. The rapid metabolism of these compounds can
lead to a short half-life, necessitating frequent dosing to maintain therapeutic levels.
Metabolites generated during this process may exhibit altered pharmacological activities,
potentially influencing the overall therapeutic effect. Understanding the metabolic pathways
of phytochemicals is crucial for predicting their in vivo behavior and optimizing drug develop-
ment strategies [81].
Elimination, primarily through renal excretion, contributes to the overall pharmacokinetics of
phytochemicals. The water solubility of compounds influences their renal clearance, and those
with low water solubility may undergo enterohepatic recycling, further complicating their elimi-
nation kinetics. The rate of elimination affects the duration of action and the dosing frequency
required for maintaining therapeutic levels [82].
To address these pharmacokinetic limitations, researchers in drug discovery have explored
various strategies to enhance the pharmaceutical properties of phytochemicals. Nanoparticle-
based drug delivery systems, including liposomes, micelles, and polymeric nanoparticles, have
shown promise in improving solubility, protecting against degradation, and facilitating con-
trolled release. These nanocarriers can enhance the bioavailability of phytochemicals by over-
coming challenges related to their poor water solubility and promoting their transport across
biological barriers [83].
Prodrug design represents another approach to mitigate pharmacokinetic challenges. Prodrugs
are biologically inactive precursors that undergo enzymatic or chemical transformation in vivo to
release the active drug. This strategy aims to improve the solubility, stability, and absorption of
phytochemicals, ultimately optimizing their pharmacokinetic profile. By modifying the chemical
structure of phytochemicals, prodrugs can enhance their pharmacological properties and address
issues such as rapid metabolism [84].
While phytochemicals hold immense promise in drug discovery, their pharmacokinetic limita-
tions pose substantial hurdles. Overcoming these challenges requires a multidisciplinary approach,
integrating innovative drug delivery systems, prodrug strategies, and formulation approaches. The
ongoing research in this field aims to unlock the full therapeutic potential of phytochemicals, pav-
ing the way for their successful integration into mainstream pharmaceuticals and contributing to
improved healthcare outcomes.
        300

13.8 Quality by Design (QbD) Approaches

In the early 20th century, Sir Ronald Fisher revolutionized research methodology by advocating for
the integration of statistical analysis at the planning stages of experiments, a departure from the
traditional practice of applying statistics retrospectively. This proactive approach, aligning with
Deming’s profound knowledge principles encompassing system thinking, variation understand-
ing, theory of knowledge, and psychology, not only enhances the quality of the end product but
also establishes a foundation for robust research [85]. Interestingly, the pharmaceutical industry
lagged behind other sectors in embracing these progressive paradigms. Historically, the industry’s
focus centered on blockbuster drugs, and the formulation development process primarily relied on
One Factor At a Time (OFAT) studies [86]. This approach contrasted with the contemporary meth-
odologies of QbD and modern engineering-based manufacturing. Shifting toward these advanced
strategies represents a crucial evolution for the pharmaceutical sector, aligning it more closely with
the comprehensive and forward-thinking frameworks [87].
The integration of computational tools within QbD strategies has revolutionized the landscape
of drug discovery, offering a sophisticated and systematic approach to optimize the development
process. Computational tools, encompassing molecular modeling, bioinformatics, and artificial
intelligence, play a pivotal role in the early stages of drug discovery by facilitating a deeper under-
standing of complex biological systems [88]. One key aspect of this integration is the predictive
power of computational modeling in elucidating the structure–activity relationships (SARs) of
potential drug candidates. Molecular docking and dynamics simulations allow researchers to
explore the binding interactions between drugs and their target proteins, predicting the pharmaco-
logical effects and optimizing drug design for enhanced efficacy. In addition, QSAR analyses ena-
ble the systematic evaluation of the impact of various molecular features on drug activity, guiding
the selection of lead compounds [89].
Computational tools also contribute significantly to the identification of potential drug targets
through systems biology and network pharmacology approaches. By integrating diverse data
sources, such as genomics, proteomics, and chemical databases, these tools help unravel complex
biological pathways and identify key nodes for therapeutic intervention. This holistic understand-
ing of the biological landscape aids in the selection of targets that are not only biologically relevant
but also druggable [90].
Furthermore, the utilization of artificial intelligence (AI) and ML algorithms has emerged as a
game-changer in drug discovery. These algorithms can analyze vast datasets, identify hidden patterns,
and predict potential drug candidates with specific therapeutic properties. In the context of QbD, AI
and ML models contribute to the systematic exploration of formulation parameters, predict drug–
drug interactions, and assess the impact of different variables on product quality [91]. This accelerates
the identification of optimal drug formulations, streamlining the drug development process.
The incorporation of computational tools within QbD strategies also extends to pharmacoki-
netic and pharmacodynamic (PK/PD) modeling. These tools enable the prediction of drug ADME
properties, guiding the selection of compounds with favorable pharmacokinetic profiles. By inte-
grating PK/PD modeling into QbD, researchers can establish design spaces that ensure a balance
between therapeutic efficacy and safety [92].
Moreover, the use of computational tools in toxicity prediction and risk assessment contributes
to the proactive identification of potential safety concerns during the early stages of drug develop-
ment [93]. This aligns with the risk-based approach advocated by QbD, allowing for the identifica-
tion and mitigation of risks associated with drug candidates, ultimately leading to the development
of safer and more effective pharmaceuticals.
13.8 uality by Design ( bD) Approaches 301

13.8.1 Use of Computational Models for Formulation Optimization

The pharmaceutical sector, characterized by its process-centric and quality-oriented approach,
would conventionally be anticipated to promptly assimilate the aforementioned paradigms follow-
ing their introduction. Contrary to this expectation, however, QbD was formally recommended by
regulatory authorities such as the FDA and EMA at the onset of the new millennium. This endorse-
ment reflects a discerning acknowledgment that quality cannot be ascribed to products through
testing alone; rather, it necessitates a proactive integration during the design phase. In essence, the
regulatory authorities underscored the imperative for the pharmaceutical industry to embrace a
paradigm shift wherein quality is not an outcome of testing protocols but is intricately woven into
the very fabric of the design process itself. This directive emphasizes a strategic and preemptive
approach to quality assurance, emphasizing the necessity to embed quality principles in the early
stages of product development and design [94]. This is a direct yet late acknowledgment of the
significance of quality theory in pharmaceutical development, as briefly presented. The design of
experiments is the main component of the statistical toolbox to deploy QbD in both research and
industrial settings [95]. It is imperative to acknowledge that a multitude of mathematical modeling
techniques exists to address pharmaceutical development, particularly within the QbD and pro-
cess analytical technology (PAT) framework. One illustrative example is the application of multi-
variate data analysis (MVDA) techniques, which primarily concentrate on historical data. MVDA
can serve as an initial point of departure for the design of experiments (DoEs). Consequently, the
integration of MVDA in conjunction with DoE proves instrumental in the analysis of both nonde-
signed and designed factors. This strategic combination of techniques allows for a comprehensive
approach to understanding and optimizing pharmaceutical processes, aligning with the principles
of QbD and PAT. The utilization of these mathematical modeling tools contributes to a more
informed and systematic exploration of the complex interplay between various factors, ultimately
enhancing the efficiency and efficacy of pharmaceutical development processes [85]. Nevertheless,
in light of the proactive ethos inherent in QbD, DoE emerges as the primary systematic approach
necessitating the incorporation of statistical thinking at the inception of pharmaceutical
development – a principle that resonates with the central tenets of Fisher’s legacy. The advent of
the ICH Q8 guideline has notably fostered a conducive environment for the application of
DoE. Consequently, a discernible surge in relevant scientific research and industrial implementa-
tion has been observed. This trend has been further bolstered by the introduction of user-friendly
software tools, which streamline the construction and analysis of experimental designs, thereby
facilitating the seamless integration of DoE in pharmaceutical development processes. The
combination of QbD principles, the ICH Q8 guideline, and accessible software resources collectively
underscores the increasing significance and widespread adoption of DoE as a pivotal methodologi-
cal framework in the pharmaceutical industry [85].
QbD is a systematic development approach that commences with predefined objectives, placing
a strong emphasis on gaining a comprehensive understanding of both the product and the associ-
ated processes. This involves a meticulous focus on process control rooted in sound scientific prin-
ciples and quality risk management. The core philosophy of QbD is centered around a proactive
and methodical strategy, ensuring that quality is not just a desirable outcome but an integral part
of the entire development process. By integrating robust scientific principles and risk management
strategies, QbD facilitates a streamlined and efficient path from concept to product, fostering a
culture of continuous improvement and excellence in the pursuit of high-quality outcomes [94].
Experimental design, also known as DoE, is a meticulously structured and organized methodology
employed to discern the intricate relationships between factors influencing a given process and the
        302
subsequent output of that process. In essence, it serves as a systematic approach to acquiring pro-
found process knowledge by establishing precise mathematical relationships between the various
inputs and outputs of the process under consideration. This methodological framework provides a
comprehensive means of understanding and optimizing processes through a strategic and well-
defined exploration of the interdependencies among different factors. The utilization of experi-
mental design, or DoE, thus contributes to a more informed and scientific decision-making process
in various fields, offering a rigorous and systematic avenue for uncovering the underlying mecha-
nisms governing complex processes and facilitating advancements in process optimization and
quality enhancement [85].
A process is essentially a value-adding activity that intricately transforms a set of inputs into
outputs. This transformation is intricately influenced by various factors, which can be categorized
as either controlled or uncontrolled. The latter is aptly termed “noise” as these factors occur ran-
domly and, in the grand scheme, exert a minimal impact compared to their controlled counter-
parts. In essence, a process involves managing and optimizing these factors to enhance efficiency
and ensure that the controlled elements play a predominant role in shaping the desired out-
comes [85]. However, within the realm of experimental design, Montgomery discerns process vari-
ables by categorizing them into potential design factors and nuisance factors. These are then
further classified as controllable, uncontrollable, or noise. The latter pertains to uncontrolled and
inevitable variations, represented by experimental error. This distinction underscores the unavoid-
able nature of certain factors, emphasizing the need to differentiate them within the experimental
framework for a more nuanced understanding of their impact on the overall outcomes [86]. The
influence of controlled factors on the quality characteristics of the end product follows a consistent
pattern known as the Pareto principle. This principle, often referred to as the 20:80 rule, highlights
that a relatively small number of factors contribute significantly to the overall effect. In essence, it
posits that 20% of the causes (factors) account for 80% of the results (responses). Within the frame-
work of ICH Q8, these highly impactful factors are identified as critical process parameters (CPPs).
CPPs are defined as process parameters whose variability directly affects a critical quality attribute
of the product, underscoring their pivotal role in determining the final product’s quality [94].
Consequently, it is imperative to monitor and control CPPs to guarantee the consistent production
of the desired quality. This quality is defined by the aggregation of the product’s characteristics,
which must consistently align with predefined ranges commonly referred to as specifications.
Ensuring that these specifications are met through vigilant monitoring and control of CPPs
becomes instrumental in maintaining the overall quality integrity of the process and its final
outcomes.
In this context, ICH Q8 outlines critical quality attributes (CQAs) as essential physical, chemical,
biological, or microbiological properties that must fall within appropriate limits, ranges, or distri-
butions to uphold the desired product quality. DoE serves as an approach wherein the controlled
input factors of the process undergo systematic and purposeful variations to discern their effects
on the corresponding responses. This systematic exploration allows for a comprehensive under-
standing of the interplay between various factors and their impact on critical quality aspects, aid-
ing in the refinement and optimization of the overall production process [86].
DoE surpasses OFAT in several aspects. Notably, DoE excels in maximizing process knowledge
while minimizing resource usage and delivering accurate information efficiently. Its strength lies
in identifying factor interactions and characterizing the relative significance of each factor, ena-
bling predictions of process behavior within the design space [96]. In addition, DoE establishes a
robust cause-and-effect relationship between CPPs and quality CQAs, facilitating the optimization
of CQAs through appropriate CPP settings and defining a proven acceptable range (PAR)