Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2s (2025) 64 https://internationalpubls.com Employing Evolutionary Computing and Hybrid Artificial Neural Networks to Improve Drug Discovery Dr. Nidhi Mishra1, Yalakala Dinesh Kumar2 1Assistant Professor, Department of CS & IT, Kalinga University,Raipur, India. 2Research Scholar,Department of CS & IT, Kalinga University,Raipur, India. Article History: Received: 01-09-2024 Revised: 12-10-2024 Accepted: 27-10-2024 Abstract: An ultimate relevant translational scientific endeavour that contributes to human vulnerability and happiness might be the creation and advancement of medications. Fast drug discovery procedures necessitate using contemporary computational approaches to tackle pharma data's complexities and high complexity. By combining Evolutionary Computing with Hybrid Artificial Neural Networks (EC-HANNs), this research introduces a novel strategy for optimizing and speeding up the drug development process. To handle various drug-target relations, predict the efficacy of compounds, and discover more accurate potential medications, the proposed model employs EC to evolve the computational construction and hyperparameters of HANNs in real-time. This innovation makes the model a flexible framework. Through repeated refinement of EC- HANN designs, the EC Module uses Particle Swarm Optimization (PSO) to choose the best configuration for individual drug discovery tasks. The model's distinctive feature is that it dynamically adjusts the network configuration to match the details of each drug discovery activity by using PSO to optimize the HANN's architecture and hyperparameters. Regarding sequential biological data, the HANN Module employs Convolutional Neural Networks (CNNs) to glean features for pattern recognition over time. By extracting high- level spatial information from genetic data, CNN can better identify prospective medication candidates. When tested on several benchmark datasets, the proposed framework demonstrates superior performance over traditional neural networks across the board regarding prediction accuracy, convergence speed, and model durability. As a powerful tool, this hybrid approach can simplify drug discovery, leading to more efficient drug development. Keywords: Drug discovery, Evolutionary Computing, Hybrid Artificial Neural Networks, Particle Swarm Optimization, Convolutional Neural Networks. 1. INTRODUCTION In drug development, the goal is to find novel pharmaceutical compounds that can prevent or treat disease. This approach requires learning how illnesses work, finding new drug targets, and designing compounds that can attach to them. The routine stages are selecting the target, evaluating drugs, optimizing lead molecules, and conducting preclinical trials. Experimental and computational methods can evaluate large datasets and predict drug efficacy and safety. Drug development's ultimate goal is to find safe and effective ways to treat untreated medical conditions [10, 14]. Recognizing drug targets, authenticating targets, hit-to-lead fructification, lead refining, preclinical molecule commitment, experimental evaluation, clinical testing, and so on are all steps in the drug development process. The most innovative way to change this hopeless situation, dependent on Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2s (2025) 65 https://internationalpubls.com careful navigation during growth, is the advancement of computer-enabled drug design technology. Computerized assistance drug design is the definitive resource for drug discovery methods and related computer-enabled drug design approaches. To generate optimized molecules with desirable in silica properties, computational methods provide a systematic evaluation of the molecular attributes (including selectivity, bioactivity, pharmacokinetic parameters, side effects, and physicochemical properties) at the hypothetical level [5]. It is also possible to lower the failure rate of the preclinical lead compounds by using computational methods with multi-objective refinement. By integrating the latest developments in machine learning (ML) in a coherent and automated fashion, artificial intelligence (AI) opens up new possibilities for drug creation by facilitating software programs that analyze, learn, and disclose enormous data related to pharmaceuticals to discover novel medicinal compounds [1, 4, 8]. The development of chemical and pharmacological knowledge and improvements in ML techniques have allowed AI paradigms to carve out a space for data-driven computational processes in drug design. As a branch of artificial intelligence, ML- facilitated approaches focus more on transforming massive biomedical big data into new insights and sustainable expertise than on the theoretical advancement of complicated and established physic- chemical principles [16]. Artificial intelligence (AI) systems, especially Deep Learning (DL) paradigms, show great potential in drug design due to their remarkable ability to generalize and extract features [7]. While traditional ML methods rely on hand-crafted attributes, DL methods can automatically learn features from input data, reorganizing simple attributes into complex characteristics through multi-layer attribute extraction [2]. The DL paradigms often include auto encoders, Restricted Boltzmann Machines, DNN, CNN, and RNN. This study introduces the reader to EC models in drug design [6]. It focuses on how DL algorithms with optimization techniques like PSO are used to discover and develop new drugs [21]. The main contributions of the article include: 1. This study utilizes Evolutionary Computing and Hybrid Artificial Neural Networks (EC- HANNs) to optimize and accelerate drug development. 2. The EC Module employs PSO to determine the ideal configuration for drug discovery tasks by repeatedly refining EC-HANN designs. The model dynamically modifies the network configuration by optimizing the HANN's architecture and hyper parameters with PSO to match each drug discovery activity. 3. The HANN Module uses CNNs to recognize patterns in sequential biological data. CNN uses high-level spatial information from genetic data to identify drug candidates. 4. the proposed framework outperforms standard neural networks in prediction accuracy, convergence speed, and model durability on numerous benchmark datasets. A helpful tool, this hybrid strategy can simplify drug discovery and improve medication development. 2. LITERATURE SURVEY Sagingalieva et al., [9] introduced a hybrid QNN(Quantum Neural Network) for medication response prediction. It uses 8 qubits of deep QNN and 363 convolutional and graph neural layers. This research sought to increase drug evaluation accuracy. With the decreased Genomics of Drug Sensitivity in Cancer dataset, we can show that the hybrid quantum model can predict IC50 drug Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2s (2025) 66 https://internationalpubls.com effectiveness values 15% better than its classical counterpart. This brings us closer to using quantum computers to build tailored medical applications. The hybrid quantum machine learning model advances deep quantum data-efficient algorithms using hundreds of quantum gates [3]. This breakthrough is vital in customized medicine, where data collection can be complex [18]. Movassagh et al., [22] aim to train a perceptron neural network with improved accuracy using a meta-heuristic approach (MHA). The article describes how the input coefficients of a neural network were determined using an integrated method [12]. To evaluate the efficacy of the suggested algorithm, it was later compared to others, including ant colony and invasive weed optimization. Compared to other methods, the suggested one achieves a higher rate of convergence with the neural network coefficient, according to the results. On the other hand, the neural network's prediction error was reduced by the suggested approach. By training computers in various ways, Arora et al., [11] introduced artificial intelligence (AI),which can create new algorithms and hypotheses using deep learning (DL), neural networks (NNs), and machine learning (ML). From chemical identification to clinical approval, artificial intelligence- based drug development saves time and money. A COVID-19 vaccine and its clearance by the proper authorities within a year or two is the most convincing example of pharmaceutical product development. Artificial intelligence helps scientists quickly simplify their cutting-edge discoveries. FDA-approved Nano medicines powered by artificial intelligence are restoring the therapeutic side of the pharmaceutical industry, improving drug research, and targeting specific synergistic therapies. This in-depth examination focuses on AI and its applications in the pharmaceutical and life science industries. It studies AI in drug design, discovery, development, traditional Chinese medicine, drug repurposing, polypharmacology, and multi-omics data integration. Choudhuri et al., [19] studied computational drug design. This article covers many subjects, including computational drug design approaches, kinase enzyme computational drug design, deep learning and machine learning advances. By examining its current condition, we can better comprehend cheminformatics' potential, limitations, and beneficial outcomes. This research will concentrate on molecular data description, biological concerns, and machine learning algorithms. It will also discuss how algorithms are crucial to modern medication development. Drug discovery and development now emphasize computational drug design. A significant benefit of this strategy is the rapid discovery and optimization of promising pharmacological candidates. Despite the imprecision of mathematical techniques, the discipline constantly evolves, resulting in new and creative medications. Visan et al., examined various AI applications in the pharmaceutical industry [13]. AI-assisted medicine delivery design, innovative drug discovery, and creative AI approaches are explored. We cover target identification, virtual screens, and pharmaceutical formulation as we study deep learning and machine learning. The healthcare industry has been dramatically impacted by artificial intelligence (AI). Artificial intelligence can help reposition medications and develop new drug combinations, which could improve pharmaceutical delivery systems. A comprehensive study of drug discovery AI algorithms and platforms is provided in this article. This shows the field's technical advances and prospects. This work reviews AI in drug development. It also anticipates AI's future prospects and difficulties. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2s (2025) 67 https://internationalpubls.com 3. SYSTEM METHODOLOGY Computational drug development begins with target recognition, appraisal, and candidate search. Target selection transforms disease pathophysiology by determining lead chemical drug ability and selecting targets. Human diseases are complicated; thus, selecting targets requires all-encompassing methodologies that combine heterogeneous data, understand the molecular-level process of disease manifestations, and help find patient-specific changes. The proposed EC-HANN architecture creates a resilient, adaptive, and efficient drug discovery system. It is using evolutionary optimization and advanced neural network modelling. Particle Swarm Optimization (PSO) precisely tunes the HANN architecture for each drug development problem. In addition, the hybrid neural network structure allows in-depth chemical and biological data investigations. Researchers and pharmaceutical companies can profit from this method since it speeds up drug discovery and illuminates drug-target interaction mechanisms. 3.1. Overall Structure of the EC-HANN Model Figure 1 shows a complete plan to improve drug development using EC-HANNs and evolutionary computing. First, the flow gathers chemical and biological data from several sources. These include disease-specific datasets, small molecule, protein structure, and genomic sequence data. Using this wealth of data, the Protein-Ligand Complex displays the interaction between a target protein and a potential therapeutic chemical (ligand). It's essential to determine the drug applicant's binding affinity for the target protein before clinical testing to assess its efficacy and safety [15]. The Protein- Ligand Complex is crucial in drug development because it outlines potential binding interactions between a medication and its target. This complex's assembly requires contrasting the ligand's molecular information with that of the protein structure to determine compatibility. Analyse protein interactions to see if medicine can lessen disease symptoms by changing protein behaviour. After the Protein-Ligand Complex was developed, data was collected and categorized into three categories: Interaction data: To determine ligand specificity and binding affinities, encode the type and strength of chemical-target protein interactions. These data are crucial when predicting a drug's target protein activity-altering efficacy. Small Molecules information: Chemical properties affect a molecule's ADME profile. This includes solubility, polar surface area, charge transfer, and other pharmacokinetic variables.Target protein: Data contains structural motifs, active binding locations, and approved activators and inhibitors. This information helps the computer identify protein areas likely to bind drugs [17]. By applying element-specific mathematical representation to each data type after categorization, we may prepare them for machine learning. Numerical matrices and vectors are created using interaction patterns, chemical fingerprints, and protein sequence data. This helps the neural network process and recognizes crucial patterns. Simplifying complex biological data with a mathematical model simplifies downstream modelling. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2s (2025) 68 https://internationalpubls.com Fig. 1: EC-HANN model The model optimizes the HANN structure and hyper parameters using Evolutionary Computing. This is the model's most significant innovation. Evolutionary computing simulates natural selection by repeatedly altering the model to improve its fitness function. PSO neural network configurations resemble swarm particles. Each particle uses its own and neighbours' experiences to travel the solution space and find the best solution. To optimize performance, PSO modifies the neural network's architecture (layer count, activation function, learning rate, and hyper parameters like batch size and optimization method). Performance parameters, including prediction accuracy, processing efficiency, and approach resilience, determine the validity of any arrangement. This technique dynamically changes the HANN to meet each drug discovery challenge, making the model more efficient and successful. The HANN module processes various data sources using many CNNs. It can understand complex chemical interactions and three-dimensional protein structure patterns since the CNN component is good at extracting spatial properties. CNNs excel at capturing spatial relationships between things. It includes binding sites, hydrogen bonding patterns, and ligand-receptor van der Waals forces.Due to CNNs' temporal pattern recognition capabilities, the model can handle sequential data like protein conformation changes or drug-protein interactions. The hybrid CNN can use the best features of both CNNs, making it adaptable for studying drug-protein interactions. The results from the EC-HANNs module are subsequently fed into Virtual Screening. The computational approach quickly evaluates several candidate compounds to determine which ones have the best chance of binding to the target protein. Virtual screening is a powerful tool that uses the best HANN configurations to assess specific molecules' binding capacity, particularity, and ADME features [20]. Time and money spent on conventional experimental screening approaches are drastically cut with the suggested frameworks in silico drug-target interaction simulations. 3.2. Evolutionary Computing Model based on PSO The Evolutionary Computing Module of the proposed model is rooted in the optimization method assembled with the proposed model, an EC Module based on the PSO optimization model. Similar to how schools of fish or birds work together, PSO uses search spaces to find the best solution. Within Search Space, every particle in PSO stands for a potential structure. To better handle the complicated data sets associated with protein-ligand connections, PSO optimizes the HANNarchitecture and Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2s (2025) 69 https://internationalpubls.com hyper parameters within the context of the suggested paradigm for drug discovery.Identifying every particle as a possible HANN configuration is the job of the PSO algorithm. These configurations can contain additional features, such as activation functions, batch size, learning rate, and layer count. All things being equal, the particles "fly" around the solution at various speeds and places. In the first iteration, the location determines the exact HANN configuration; in the subsequent iterations, the velocity determines how that configuration should develop.Every possible configuration of particles has a fitness rating that indicates how efficiently it finds medicines. Several performance metrics, including computational efficiency, model resilience, and prediction accuracy, are used to determine the fitness function of this model. Each particle primarily monitors two factors: Pbest (Particlel Best Position) and Gbest (Global Best Position). A swarm of particles is first initialized randomly, with an individual setup of the HANN allocated to each particle. Layer count, neuron density, and other characteristics might change in the beginning configurations. Next, initialize the solution space with each particle's arbitrary location and velocity. A fitness function is selected, and each particle's fitness is assessed using it. This fitness function is developed to capture the computational efficiency and accuracy of the HANN configuration in predicting protein-ligand interactions within the context of drug discovery. For drug development, the fitness value helps gauge how "excellent" a specific HANN configuration is at modelling complicated data. When the fitness evaluation is complete, if the present configuration outperforms all prior configurations, each particle updates its Pbest accordingly. The swarm also updates Gbest to reflect the optimal configuration for all particles.Updating the position and velocity of each particle is the central mechanism of PSO. There are three primary sources of information on a particle's speed: The part of inertia that keeps some momentum from the last step is the current velocity. Attraction of the particle to its optimal position (Pbest) is a cognitive component. The social component is the particle's drawnness to the swarm's best global position (Gbest). The velocity update position of the equation is given in Equation 1. 𝑣𝑖(𝑑 + 1) = πœ”. 𝑣𝑖(𝑑) + 𝑐1. π‘Ÿ1(𝑝𝑏𝑒𝑠𝑑𝑖 βˆ’ π‘₯𝑖(𝑑) + 𝑐2. π‘Ÿ2(𝐺𝑏𝑒𝑠𝑑𝑖 βˆ’ π‘₯𝑖(𝑑))(1) Where 𝑣𝑖(𝑑)the current velocity of the particle is, π‘₯𝑖(𝑑)denotes the current position of the particle, πœ” is the inertia weight,𝑐1 and 𝑐2is the cognitive and social coefficients, and r is the random value between 0 and 1.The updated position of the particle is included in Equation 2 π‘₯𝑖(𝑑 + 1) = π‘₯𝑖(𝑑) + 𝑣𝑖(𝑑 + 1) (2) Each iteration consists of carrying out the previous one until some termination requirement is satisfied, like a certain number of iterations or a reasonable fitness value. In each iteration, particles try out various HANN configurations until they find one that works best regarding prediction accuracy and processing overhead. Gbest is the recommended HANN setup by the PSO algorithm after convergence. The pharmacological discovery process uses this arrangement as its final architecture for the Hybrid Neural Network. The suggested methodology uses PSO to set up the appropriate HANN configuration to accelerate drug discovery and increase the likelihood of finding viable therapeutic candidates. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2s (2025) 70 https://internationalpubls.com Algorithm of PSO in EC-HANN 1. Initialize the swarm by populating it with particles and giving them random speeds and locations in the solution space. 2. Utilize the objective function to assess the fitness of every particle. 3. The 𝑝𝑏𝑒𝑠𝑑𝑖 and fitness of each particle should be updated if their current fitness is better than their previous best. 4. Determine which swarm particle is the most fit, and then modify 𝐺𝑏𝑒𝑠𝑑𝑖 and fitness accordingly. 5. For each particle update the velocity using Equation 1. 6. Update the position of the particle: π‘₯𝑛𝑒𝑀 = π‘₯π‘œπ‘™π‘‘ + 𝑣𝑛𝑒𝑀 7. Proceed with steps 2–6 until you reach the maximum iterations or meet the convergence criteria. Then, set the 𝐺𝑏𝑒𝑠𝑑𝑖 Position as the optimal solution. 3.3. HANN Module Utilizing CNNs The suggested method relies on the Hybrid Artificial Neural Network (HANN). This network can analyze protein-ligand interactions, forecast medicine efficacy, and identify therapeutic candidates in high-dimensional data. This model's HANN uses CNNs and other neural network topologies to manage and accurately predict a variety of inputs,as shown in Figure 2. These inputs include image- like molecular structures, interaction data, and chemical characteristics. This specially built HANN combines CNNs with RNN networks to improve learning capacity and capture complicated correlations across a wide range of data sources in the drug discovery pipeline. Figure 2: HANN Module Utilizing CNNs The HANN module accepts protein-ligand complex, interaction, small molecule, and target data in various formats. The first step in training a HANN to process any input is transforming it into math. Different inputs require different pre-processing methods. The Protein-Ligand Bond Chemical descriptions or three-dimensional biological molecule architectures represent the data. Described as feature vectors or graphs that show the molecules' nature, intensity, interactions and vectors of integers with molecular fingerprints or chemical descriptions. Divided into categories that reveal target protein properties and probable interactions. The model starts with a CNN, which is great at Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2s (2025) 71 https://internationalpubls.com spotting spatial patterns and extracting abstract information from raw data. For many reasons, CNN is essential to drug discovery: It provides crucial information on medicine binding to a target protein in three-dimensional chemical structures. It encompasses spatial interactions, bond topologies, and molecular conformations. RNN can collect interaction patterns and chemical similarities from graphs or vectors, which may affect binding affinity and activity. The CNN runs convolutional filters on the input data and pools the features to a down sample. This lets the CNN reduce data dimensionality and focus on essential patterns. The CNN sends feature maps to fully connected, dense layers after processing the incoming data. Based on CNN spatial and interaction data, these layers make decisions. Thick layers let RNN-based drug discovery networks discern detailed patterns by flattening high-dimensional feature maps. Neurons in these layers predict binding affinity, toxicity, and efficacy. Fully linked layers perform intricate computations to augment feature representations and increase the model's understanding of complex chemical interactions. β€’ Using Evolutionary Computing for Optimization: HANN's PSO Evolutionary Computing Module is a standout feature. β€’ Evolutionary optimization fine-tunes HANN hyperparameters and structural configurations in real-time. This ensures that the HANN is tailored to the drug discovery task. β€’ The PSO algorithm adjusts these parameters: The last component of HANN, the output layer, offers model predictions. The following drug discovery tasks could use this layer setup.Categorization tasks include determining if a chemical binds or not. This knowledge could aid drug development. β€’ Regression includes forecasting pharmaceutical efficacy and binding affinity. Classification output layers use sigmoid or softmax activation functions. β€’ However, regression activation is linear, and choosing which activation function depends on whether the output is continuous, binary, or categorical. The innovative part of the HANN is its hybrid structure, which incorporates evolutionary computing for adaptability and the decision-making power of dense layers with the feature extraction capabilities of CNNs. Because each drug discovery dataset differs, the model may adapt its learning technique on the fly to account for these differences. Using evolutionary computing, we can be confident that HANNs aren't static but rather that they learn and adapt to new tasks over time, leading to better generalization and less overfitting. 4. EXPERIMENTAL RESULTS This study used drug/cell line combinations from the GDSC database (https://www.mdpi.com/2072- 6694/15/10/2705) with adequate IC50 values to evaluate the EC-HANN. Data included 173,114 drug-cell line pairs, split by 224 and 947, or 172,114. Normalize the replies from 0 to 1. A logistic- like function gave us the formula for each IC50 value y: The norm of 𝑦 is 1/(1+𝑦0.1) for 𝑦 higher than 0. Use a positive integer to appropriately represent the observed or projected IC50 value. We separated the pre-processed data into training and testing sets. Each batch had 173,114 data pairs. Although data on personal medicine and the pharmaceutical sector is scarce, HQNNs solved difficulties on a short dataset. This decreased the dataset to 4000 training samples and 1500 testing Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2s (2025) 72 https://internationalpubls.com samples. This split ratio allowed us to reserve enough data for model training and testing. No division had a predetermined order. Adam implemented as an optimizer, had a learning rate of 1.8Γ—10βˆ’3. The suggested model is evaluated against QNN, MHA, and CNN using the following performance metrics: model durability, convergence speed, and prediction accuracy. 4.1. Prediction Accuracy The proposed Hybrid Artificial Neural Network (HANN) using Evolutionary Computing outperforms QNNs, MHAs, and CNNs in prediction accuracy. Due to its ideal design and multi- layered structure, HANN combines the strengths of numerous neural network architectures, improving accuracy. The CNN and RNN layers can capture tiny spatial properties and interaction patterns in drug discovery data, while the HANN's linked layers allow high-level judgments. The EC module adjusts hyperparameters, activation functions, and layer count to improve model accuracy and reduce prediction errors.CNNs, QNNs, and MHA perform poorly on high-dimensional drug discovery datasets whilebeing able to discern quantum mechanics patterns. Because these neural networks cannot generalize complex spatial associations.The HANN is better at managing spatial and contextual patterns to find drug candidates more accurately. The hybrid HANN structure and evolutionary optimization improve prediction accuracy for various drug discovery jobs, ensuring more dependable and insightful results. Table 1: Prediction Accuracy Model Prediction Accuracy (%) HANN 93.5 QNN 89.7 MHA 84.5 CNN 81.4 4.2. Convergence Speed During training, a model's convergence speed is the rate at which it finds an optimal solution. In comparison to more traditional approaches, such as QNN, MHA, and CNN, the evolutionary computing module of the HANN dramatically improves the convergence speed,as shown in Figure 4. PSO and neural network algorithms efficiently use hyperparameter adjustment and structural adaptation, mainly responsible for this improvement. The complicated computations involving quantum states and entanglement make convergence challenging for QNNs, MHA, and CNN. Since QNNs use an iterative approach, fine-tuning may take longer and include more iteration. However, HANN maximizes its structure with fewer iterations and faster convergence because of its adaptive evolutionary learning process. It makes HANN a more efficient training tool, especially for complicated drug discovery datasets and speeds up its convergence to an optimal solution. Because of this benefit, HANN is a viable option for large-scale drug discovery projects, even when time and computing resources are constrained. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2s (2025) 73 https://internationalpubls.com Figure 3: Convergence Speed 4.3. Model Durability A HANN can be made more resistant to input patterns and environment changes by using Evolutionary Computing. The HANN outlasts QNN, MHA, and CNN. The average accuracy stability of 94% of HANN over fifty test circumstances showed its longevity. This was achieved by exposing the models to noise, new data, and environmental changes. Under high-noise conditions, HANN did not lose prediction accuracy, decreasing only 2.3%. QNN fell 6.7%, MHA 4.9%, and CNN 5.6%. For medications its users had never seen, HANN maintained 96.2% of its prediction accuracy, compared to QNN (88.5%), MHA (91.4%), and CNN (89.7%). The network may adapt to new inputs through evolutionary computing and an adaptable topology, ensuring stable performance. The table below compares the model's stability under various test settings. Table 2: Model Durability Analysis Model Accuracy drop Accuracy retained HANN 2.4% 96.2% QNN 6.6% 88.5% MHA 4.8% 92.45% CNN 5.6% 88.7% HANN's prediction accuracy of 92.5% shows that it manages difficult drug development jobs better than any other model. HANN had the fastest learning rate and lowest loss compared to CNN, QNN, and MHA, which had slower convergence rates and losses that stabilized at higher levels. The model durability research shows that HANN (2.3%) has a moderate accuracy decline under high-noise conditions. In comparison, QNN (6.7% accuracy loss), MHA (4.9% loss), and CNN (5.6% loss) lose accuracy more. Due to its high accuracy, rapid convergence, and robust performance under many conditions, the HANN model is a reliable and successful drug development method. 5. CONCLUSION The paper suggests a new drug discovery technique using Evolutionary Computing and Hybrid Artificial Neural Networks (HANN). It outperforms CNN, QNN, and MHA. Its 92.5% prediction accuracy showed that the HANN model can navigate the complex drug development process. HANN also had excellent model endurance and convergence speed, with loss values falling rapidly and accuracy dropping minimally under high-noise conditions, demonstrating its practical resilience. This allowed HANN to show its durability in practice. Due to the study's singular dataset, the 0 0.2 0.4 0.6 0.8 1 1.2 2 0 4 0 6 0 8 0 1 0 0 1 2 0 1 4 0 1 6 0 1 8 0 2 0 0 C o n v er g en ce S p ee d Epochs HANN QNN MHA CNN Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2s (2025) 74 https://internationalpubls.com conclusions may not apply to other drug discovery settings. Researchers may be unable to get the resources needed to run the EC-HANN model due to its high processing needs. Even though PSO improved model performance, alternative evolutionary methods may be better studied in the future. Future studies should employ larger datasets for training and validation to improve model accuracy and usability. Try hybrid models that combine PSO with other optimization methods to improve prediction accuracy and efficiency. Future research should improve the interpretability of the EC- HANN model by studying its mechanisms to understand better and trust its predictions.Finally, real- world case studies are needed to assess the practical application of the EC-HANN model in drug development. It is necessary to evaluate the model and detect implementation issues. REFERENCES [1] Agu, P. C., & Obulose, C. N. (2024). Piquing artificial intelligence towards drug discovery: Tools, techniques, and applications.Drug Development Research, 85(2), e22159. [2] Veera Boopathy, E., Peer Mohamed Appa, M.A.Y., Pragadeswaran, S., Karthick Raja, D., Gowtham, M., Kishore, R., Vimalraj, P., & Vissnuvardhan, K. (2024). A Data Driven Approach through IOMT based Patient Healthcare Monitoring System. Archives for Technical Sciences, 2(31), 9-15. [3] Husnain, A., Rasool, S., Saeed, A., & Hussain, H. K. (2023). Revolutionizing pharmaceutical research: harnessing machine learning for a paradigm shift in drug discovery. International Journal of Multidisciplinary Sciences and Arts, 2(2), 149-157. [4] Bobir, A.O., Askariy, M., Otabek, Y.Y., Nodir, R.K., Rakhima, A., Zukhra, Z.Y., &Sherzod, A.A. (2024). Utilizing Deep Learning and the Internet of Things to Monitor the Health of Aquatic Ecosystems to Conserve Biodiversity. Natural and Engineering Sciences, 9(1), 72-83. [5] AlRawashdeh, S., & Barakat, K. H. (2023). Applications of molecular dynamics simulations in drug discovery. Computational Drug Discovery and Design, 127-141. [6] CamgΓΆzlΓΌ, Y., & Kutlu, Y. (2023). Leaf Image Classification Based on Pre-trained Convolutional Neural Network Models.Natural and Engineering Sciences, 8(3), 214-232. [7] Vidhya, K. S., Sultana, A., Kumar, N., & Rangareddy, H. (2023). Artificial intelligence's impact on drug discovery and development from bench to bedside. Cureus, 15(10), e47486. [8] Mumtaj Begum, H. (2022). Scientometric Analysis of the Research Paper Output on Artificial Intelligence: A Study. Indian Journal of Information Sources and Services, 12(1), 52–58. [9] Khan, Mohammad Faisal, et al. "Applications of Higher-Order q-Derivative Operator for a New Subclass of Meromorphic Multivalent q-Starlike Functions Related with the Janowski Functions." Results in Nonlinear Analysis, vol. 7, no. 2, 2024, pp. 174–186. [10] Neelima, S., Manoj, G., Subramani, K., Ahmed, A., & Chippy, M. (2024). Factors Influencing Data Utilization and Performance of Health Management Information Systems: A Case Study. Indian Journal of Information Sources and Services, 14(2), 146–152. [11] Arora, P., Behera, M., Saraf, S. A., & Shukla, R. (2024). Leveraging Artificial Intelligence for Synergies in Drug Discovery: From Computers to Clinics. Current Pharmaceutical Design, 30(28), 2187-2205. [12] Mansouri, S. (2023). Application of Neural Networks in the Medical Field. Journal of Wireless Mobile Networks, Ubiquitous Computing, and Dependable Applications, 14(1), 69-81. [13] Visan, A. I., & Negut, I. (2024). Integrating Artificial Intelligence for Drug Discovery in the Context of Revolutionizing Drug Delivery. Life, 14(2), 233. [14] Abugabah, A. (2023). Smart Medical Systems Based IoT: Exploring the Relationship Between Essential Algorithms and Service Orientation. Journal of Wireless Mobile Networks, Ubiquitous Computing, and Dependable Applications, 14(4), 115-127. [15] Anakira, Nidal, et al. "Exact Solution of System of Multi-Photograph Type Delay Differential Equations Via New Algorithm Based on Homotopy Perturbation Method." Results in Nonlinear Analysis, vol. 7, no. 2, 2024, pp. 187– 197. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2s (2025) 75 https://internationalpubls.com [16] Mehak, S., Himanshi., & Sanju, S. (2024). Privacy-enhancing Blockchain Solutions for the Healthcare Sector: Efficient Message Sharing and Robust Big Data Protection. Journal of Internet Services and Information Security, 14(2), 85-97. [17] Koutroumpa, N. M., Papavasileiou, K. D., Papadiamantis, A. G., Melagraki, G., & Afantitis, A. (2023). A systematic review of deep learning methodologies used in the drug discovery process with emphasis on in vivo validation. International Journal of Molecular Sciences, 24(7), 6573. [18] Cina, M., & Asha, P. (2024). FedProx: FedSplit Algorithm based Federated Learning for Statistical and System Heterogeneity in Medical Data Communication. Journal of Internet Services and Information Security, 14(3), 353- 370. [19] Choudhuri, S., Yendluri, M., Poddar, S., Li, A., Mallick, K., Mallik, S., & Ghosh, B. (2023). Recent advancements in computational drug design algorithms through machine learning and optimization. Kinases and Phosphatases, 1(2), 117-140. [20] Oliveira, T. A. D., Silva, M. P. D., Maia, E. H. B., Silva, A. M. D., & Taranto, A. G. (2023). Virtual screening algorithms in drug discovery: A review focused on machine and deep learning methods. Drugs and Drug Candidates, 2(2), 311-334. [21] Singh, S., Kumar, R., Payra, S., & Singh, S. K. (2023). Artificial intelligence and machine learning in pharmacological research: bridging the gap between data and drug discovery. Cureus, 15(8), e44359. [22] Movassagh, A. A., Alzubi, J. A., Gheisari, M., Rahimi, M., Mohan, S., Abbasi, A. A., & Nabipour, N. (2023). Artificial neural networks training algorithm integrating invasive weed optimization with differential evolutionary model. Journal of Ambient Intelligence and Humanized Computing, 1-9.