Intriguing of pharmaceutical product development processes with the help of arti�cial intelligence and deep/machine learning or arti�cial neural network

The objectives of current review are (1) to provide a historical overview of arti�cial intelligence and deep/machine learning (AI & D/ML) or Arti�cial Neural Network (ANN) (2) to update the �nancial dealings of pharma companies related to the application of AI & D/ML or ANN in drug discovery and development processes and (3) to showcase the application of AI & D/ML or ANN concept for optimization of analytical method conditions and formula of the dosage form. The optimization of analytical method conditions and formula of dosage form started with the employment of linear model such as design of experiment followed by non-linear model like AI & D/ML or ANN. Such type of linear and non-linear models blending in optimization processes nevertheless helped to suitably identify the in�uence of critical process parameters or critical material attributes on critical quality attributes. However, much of integration and understandable interpretation between the available data arised from clinical trials and the prevalence/progression of pandemic/endemic infections could potentially be ambitioned through the application of AI & D/ML or ANN.


Introduction
Producing pharmaceutical products without untoward effects but with desired qualities is not only the very basic requirement set by regulatory authorities but also indirectly decides the success of the pharma industry.The pharmaceutical product development process is a combined and coordinated work from numerous divisions of the pharma industry which usually starts from API discovery, synthesis and utilization to end with formulation development, market positioning and successful use at the consumer end.
Conventional API discovery and development process as well as optimization of analytical method and formula of dosage form routinely use the quality by testing (QbT) approach or one factor at a time (OFAT) strategy.Since the time consuming and chemical wastage in OFAT strategy or QbT approach is inevitable, the looking for a better alternative approach or strategy which shows less time consumption in conjunction with minimal chemical wastage becomes an urgent requirement.In terms of the response surface methodology (RSM) approach, the design of experiment (DoE) is being utilized effectively to optimize APIs synthetic process, analytical method development and formula optimization for nal pharmaceutical products (Rahman et al., 2021;Rahman et al., 2020).It should be added that the RSMlinked DoEis based on a linear model and therefore it won't consider the non-linear modelization concept to see the in uence of independent factors on the response variables.Thus, depending solely on the DoEfor such optimization processes may end with an erroneous conclusion and thus necessitates applying another approach (preferably of non-linear model based) which judiciously eliminates the conclusion errors noticed with the DoE-based optimization process.One such non-linear model-based approach recently introduced in pharmaceutical product development process is arti cial intelligence and deep/machine learning (AI & D/ML) or arti cial neural network (ANN).Since the information processing capacity of ANN is related to the functioning of the normal human brain, the estimation of the process parameters is being carried out on the number of trials performed by varying the composition of the excipients and processing conditions (Ghate et al., 2019).AI is a division of computer science, involved in problem resolution and in creating machines that can perform tasks which would otherwise require intelligence and human operators (Senthuraman, 2020).In simple words, AI & D/ML is a branch of computer science that deals with problem-solving through the aid of symbolic programming (Krishnaveni et al. 2019).The DL consists of a neural network of multiple layers that aim to emulate how the information processing is carried out by the human brain after understanding complicated patterns and feature interactions.In practical situations, deep learning helps to give structure to unstructured data and enables machines to learn to classify data without assistance (Abhinav & Subrahmanyam, 2019;Gilvary et al. 2019).The ML is a subset of AI utilizing algorithm models and uses statistical methods with the ability to learn with or without being explicitly programmed (Abhinav & Subrahmanyam, 2019).It should be added that the AI & D/ML or ANN is based on a non-linear concept to study the independent factor's in uence on response variables.Non-linearity refers to a massive parallel network distributed throughout that allows for approximation and real-time operation to exhibit unpredictability and random behaviour.

Historical overview on AI & D/ML or ANN
Table 1 delineatesthe historical overview on AI & D/ML or ANN in ascending order of milestone years.The ANN has a history that dates to the precomputer era although the original goal of AI & D/ML or ANN is to solve problems mostly related to biology, the extrapolation of its applicability is much more in recent years and even the intriguing nature of AI & D/ML or ANN especially in medical diagnosis principles as well as in forecasting/predicting the disease pattern in a particular region or whole world.Because rstly the AI & D/ML or ANN helps us understand the impact of increasing/decreasing disease progression vertically or horizontally on computational time.Secondly, the AI & D/ML or ANN helps us understand the situations or cases where the model ts best.Thirdly, it also explains why the certain model works better in certain environment or situation.Samson et al., (2016) described the ANN as an informationprocessing paradigm that is related to biological nervous systems i.e., the human brain.In the early 1940s (McCulloch & Pitts, 1943), researchers developed the "threshold logic" model, which encompassed a two-pronged approach to computational models of ANN.This model is directed at identifying the biological neural networks separate from ascertaining the correlation of the neural networks to arti cial intelligence.An unsupervised learning model utilizing neural plasticity and long-term potentiation has been introduced in the form of Hebbian learning (Morris, 1999).This type of learning was utilized in calculators and other computational instruments (Rochester et al., 1956;Farely & Clark, 1954).
Development of the perception in the model added another dimension to AI & D/ML or ANN by incorporating a two-layer computer network in pattern recognition through an algorithm (Rosenblatt, 1958).Particularly, the ANNs comprise a set of nodes, each of which receives a separate input, which is nally converted to output.After the introduction of a clear de nition of backpropagation (Werbos, 1975), the supervised learning method in which the ANN receives training in conjunction with optimization becomes easier to determine the loss of function during the anticipated output for each input value is known.In this way, the ANNs are linked to single or multiple algorithms to solve the problems (Paul et al., 2021).Although the AI & D/ML were coined around in the 1950s, it now becomes a slogan in pharma industries especially after nding out their bene ts in handling increased volumes of raw data following the introduction of advanced algorithms (Sharma, 2019).
[Insert Table 1 here] Pharma companies' nancial dealings related to the application of AI & D/ML or ANN in drug discovery and development processes Table 2 displays the nancial dealings of Pharma companies concerning the application part of AI & D/ML or ANN, particularly in drug discovery and development processes.The entry of AI & D/ML or ANN helps to shorten not only the new drug development period but also it signi cantly minimizes the utilization of manpower and considerable reduction in expenditure related to the API development.For example, the German-based biotechnology company, Evotec, has partnered with a UK-based company, Exscientia, for the small molecule drug discovery process.Within a short period of 8 months, the discovered small drug molecule entered Phase 1 clinical trials which might usually have taken 4-5 years to deliver the drug candidate from the traditional drug discovery process (without utilizing AI & D/ML or ANN).
[Insert Table 2 here] AI & D/ML or ANN in optimization of analytical method conditions and formula of dosage form Before entering the discussion related to optimization of analytical method conditions and formula of dosage form,it needs to be emphasized that the ANN simply mimics the principles of information processing handled by the human brain wherein the in uence of critical material attributes variation on critical analytical attributes (CAAs) can be predicted by segregating different sets of data (generated from numerous trails) into training, testing and validating (Samson et al., 2016).For this purpose, the ANN must be coupled with an algorithm to attain the "best t" optimum values for a method (Ghaheri et al., 2015).The ANN-linked algorithm produces a highly reliable and better predictor of the optimum values for a method than the RSM-based linear model (Sha & Edwards, 2007).However, the ANN relies on the number of experiments/trials conducted and consequently, it is highly likely that too high/a smaller number of trials would result in error and fault in the predictions (Ghate et al., 2019).Therefore, the ANN takes the trials of the linear model (RSM following face-centred central composite design (CCD) for generating the non-linear model [ANN-linked Levenberg-Marquardt (LM) algorithm] to predict the optimum regions for the studied CAAs (Rahman et al., 2021).Furthermore, the ANN-linked LM is a potent chemometrics method because of its high performance and good prediction for non-linear systems (Ghaedi, 2015).The typical network architecture of AI & D/ML or ANN is organized in three-different layers, viz., one input, one output and one or more hidden layers.Figure 2   needs to be integrated with the DoE approach for optimizing conditions for analytical method and formula for the dosage form.This type of integration between AI & D/ML or ANN and DoE allows the coining of new terminology called, "double-stage systematic optimization".The double-stage systematic optimization was therefore started initially by using the conventional DoEapproach and then by the application of AI & D/ML or ANN.For instance, Rahman et al., (2021) have used the RSM generated from face-centered CCD of DoE while the ANN is linked with the LM algorithm of AI & D/ML or ANN.Table 3 displays selected non-comprehensive publications showing the involvement of AI & D/ML or ANN in the optimization of analytical method conditions and the formula of the dosage form.
[Insert Figure 4 & Table 3 here] Because of AI & D/ML or ANN's advantage in dealing with complex and unstructured data, it is well suited for addressing a wide range of applications in the pharmaceutical sciences and easing up the process (Simões et al., 2020).Table 4 displays the various algorithms usage coupled with AI & D/ML or ANN in different pharmaceutical product development processes.
[Insert Table 4 here] Representing the drug release process by using computationally simple empirical models is a challenging task since there are complicated interactions between formulation and processing variables.The effort in the pre-prescription step would be considerably reduced if AI & D/ML or ANN model could forecast drug release, and the accuracy of the predictions has been proven.Nagy et al., (2019) used the near-infrared (NIR) and Raman spectra to compare four three-layer ANN models to the standard partial least square (PLS) regression to predict the dissolution pro le of extended-release anhydrous caffeine tablets.Brahima et al., (2017) used an MLP for the modelling of ribo avin release behaviour from poly(NIPA-co-AAc) hydrogels.The results showed that the function of ANN was validated, and when compared to the RSM using the mean square error (MSE), the ANN was more appropriate for predicting the release of ribo avin hydrogels and had great generality over the release behaviour of hydrogel.Additionally, Elman neural networks (ENNs) and other dynamic neural networks can also be used to forecast dissolution

Conclusion
The DoE (linear model)-supported AI & D/ML or ANN concept is currently being established to optimize not only the analytical method conditions for single or dual drug quanti cation but also the formula development for dosage form.On the other hand, additional ways are also identi ed in recent years to omit the linear model support for AI & D/ML or ANN for intriguing of pharmaceutical product development processes.To go ahead directly with non-linear models, the employment of multiple-inputmultiple-output and multiple-input-single output-based ANN architecture, self-organizing map, MLP network trained with backpropagation algorithm, three-or multi-layer feed-forward ANN, etc.   Publication title Short summary of the publication of design of experiments approach-driven arti cial intelligence and machine learning for systematic optimization of reverse phase high performance liquid chromatography method to analyze simultaneously two drugs (cyclosporin A and etodolac) in solution, human plasma, nanocapsules and emulsions (Rahman et al., 2021).
The application of arti cial neural network and least square support vector machine methods based on spectrophotometry method for the rapid simultaneous estimation of triamcinolone, neomycin and nystatin in skin ointment formulation (Abasi et  Application of the arti cial neural network to optimize the formulation of self-nanoemulsifying drug delivery system (SNEDDS) containing rosuvastatin (Vu et al., 2020).

-
The authors were able to improve the formulation of rosuvastatin SNEDDS using an arti cial neural network (Vu et al., 2020).
Modelling the absorbance of a bioactive compound in HPLC method using arti cial neural network and multilinear regression (MLR) methods (Abdullahi et al., 2020).

-
The ANN model performed marginally better than the MLR model (Abdullahi et al., 2020).
-ANN, PLS and principal component regression are three multivariate analytic approaches that have been created (Hasan et al., 2020).

Figure 1
Figure 1 shows the possible areas of AI & D/ML or ANN in the pharmacy eld of the healthcare system.
portrays the schematic architecture of AI & D/ML or ANN having three input, ten hidden and three output layers (3:10:3).The architectural structure of AI & D/ML or ANN is the most common multi-layered perceptron (MLP) type which is built on four different elements, input, hidden and output layers along with connections or weights.Interestingly, the MLP type AI & D/ML or ANN works in two phases, training and testing.The training phase is based on the iterative demonstration of the available data pattern to teach the AI & D/ML or ANN for accomplishing the designated assignment.
Figure 4 depicts the possible way to integrate AI & D/ML or ANN in the optimization of analytical method conditions and formula of the dosage form.It can be seen from Figure 4 that the AI & D/ML or ANN pro les.Petrovic et al., (2012) used both an ENN and an MLP to characterise the release curve of tablets and determined the wide applicability of ANN.Husseini et al., (2009) and Moussa et al., (2017) used ANN to optimise the ultrasonic release of APIs in preparations (such as liposomes and micelles) to keep therapeutic concentrations constant at speci c sites.Han et al., (2018) predicted the disintegrating time of disintegrating oral tablets by neural network techniques.A few selected examples of analytical method condition optimization and formula optimization by integrating the AI & D/ML or ANN concept are narrated below.SVM in formulation development Wang et al. (2022) used particle swarm optimization along with the least square support vector machine (PSO-LSSVM) to simplify the optimization process.The results of the prediction model and Taguchi design were compared with PSO-LSSVM.Additionally, this model provided lower costs and a more e cient design of pharmaceutical formulation.SVM in analytical method development Keyvan et al., (2021)suggested UV spectrophotometric method development using feed-forward arti cial neural network (FFNN) and least square support vector machine (LS-SVM) to simultaneously investigate Sofosbuvir and Daclatasvir in tablet production and biological uid.Results indicated that the technique has a high potential for predicting component concentrations in dosage forms with a shorter analysis time.GA in formulation development Kumar and Kumar (2019) used integral hybrid GA with BPANN and RSM based on a central composite design considering water fraction, surfactant fraction, powder density and ultrasonication time as analysing parameters.Results indicated that the multi-objective hybrid GA model was successful in establishing robust results compared to the conventional method.GA in analytical method developmentIn the study byAttia et al., (2021)GA-ANN was used to quantitatively analyse the UV absorption of velpatasvir and sofosbuvir which revealed some overlap, indicating di culty in the simultaneous estimation of two drugs.GA-ANN proved to be effective in estimating drugs, with acceptable values of root mean square errors for calibration and prediction.Autoencoder in analytical method developmentKensert et al. (2021)developed a deep one-dimensional convolutional autoencoder that simultaneously eliminates baseline noise and baseline drift to detect and quantify analytes in a mixture of chromatograms with high number and diversity surpassing the approaches like Savitzky-Golay smoothing, Gaussian smoothing and wavelet smoothing.AI & D/ML or ANN in optimizing eutectic solvent systemThe use of AI & ML or ANN to predict and select solvent systems is a very interesting integration of academia and industry.AI & D/ML and ANN can assist in optimizing eutectic systems by designing a solvent system based on appropriate properties.By involving AI & D/ML or ANN-based algorithms, the eutectic solvent system may be chosen by automatically separating the products from the reaction solution (self-precipitation)(Amar et al., 2019;Von Lilienfeld, 2018).

Table 1
Historical overview on arti cial intelligence & deep/machine learning or arti cial neural network in ascending order of milestone years [Partially taken from Manikiran & Prasanthi, 2019]

Table 2
Non-comprehensive nancial dealings of pharma companies related to the application of arti cial intelligence & deep/machine learning or arti cial neural network in drug discovery process [Partially taken from Savage, (2021)]

Table 3
Selected non-comprehensive publications showing the involvement of AI & D/ML or ANN in optimization of analytical method conditions and formula of dosage form

Table 4
Various algorithms usage in different pharmaceutical product development processes DBN algorithm is able to establish a classi cation system into drug like and non-drug like potential druggable candidates from ZINC datasets to