Artificial Intelligence in Anti-Inflammatory Drug Discovery

Abstract

Chronic inflammation is the underlying cause of numerous diseases, including arthritis, cardiovascular disorders, and autoimmune conditions. Traditional drug discovery for anti-inflammatory agents is a costly and time-consuming process. Artificial Intelligence (AI) and Machine Learning (ML) have emerged as transformative tools, accelerating the identification of novel targets and the optimization of lead compounds. This review explores various AI methodologies—such as Deep Learning (DL), Generative Adversarial Networks (GANs), and Graph Neural Networks (GNNs)—in the context of anti-inflammatory drug design. We discuss real-world applications, challenges in data quality, and the future of AI-driven therapeutics.

Keywords

Introduction

Feature	NSAIDs	Corticosteroids
Primary Target	COX-1 and COX-2 Enzymes	Glucocorticoid Receptors
Speed of Action	Fast (Minutes to Hours)	Moderate (Hours to Days)
Discovery Focus	Selective COX-2 inhibition to reduce GI toxicity	Reducing systemic side effects (Weight gain, Bone loss)
AI Application	Predicting binding affinity to COX enzymes	Modeling complex gene expression changes

Current anti-inflammatory drugs often have significant side effects:

• NSAIDs can cause stomach ulcers and kidney issues.
• Steroids can lead to bone density loss and immune suppression.

AI researchers are now focusing on discovering "Selective Biologics" or "Small Molecules" that target only the specific inflammatory pathway without affecting healthy organs.

1.2 Anti-Inflammatory Drugs

 Inflammation is a complex biological response of the body tissue to harmful stimuli such as pathogens, irritants, and damaged tissue. This process is extremely complicated and involves multiple types of immune cells, blood vessels, and numerous molecular biomarkers and targets. While acute inflammation is usually the initial response of the body to harmful stimuli, chronic inflammation involves a progressive shift in the type of cells present at the site of inflammation and is characterized by simultaneous destruction and healing of the tissue. While inflammation is an extremely important function of the healthy human body, inflammatory abnormalities are a large group of disorders that underlie a vast variety of human diseases such as allergic reactions, myopathies, immune system disorders, cancer, atherosclerosis, and ischemic heart disease. [18, 19] Therefore, anti-inflammatory medications are prescribed to limit the negative effects of the inflammation process in these diseases. Additionally, anti-inflammatory drugs remedy pain by reducing inflammation. Anti-inflammatory drugs are divided into two main categories: 1) corticosteroids, a class of steroid hormones and their synthetic analogues that are produced in the adrenal cortex of vertebrates (e.g., dexamethasone, and prednisone). [20] 2) Nonsteroidal anti-inflammatory drugs (NSAIDs) that counteract the cyclooxygenase (COX) enzyme, which synthesizes prostaglandins, creating inflammation (e.g., aspirin, ibuprofen, and naproxen). [21] Given the very complex nature of the inflammation processes in the body and the number of biochemical molecules involved, there likely are multiple target proteins that can be used to develop drugs to limit or inhibit inflammation and subsequently pain. This especially holds true in out era, that computer assisted methods used in bioinformatics unveil new structural and mechanistic details about the molecules involved in inflammation and their interactions, regularly. In addition, both common corticosteroid and NSAIDs have multiple side effects, intensifying the urge to invest time and effort in finding new alternative anti-inflammatory (and subsequently pain relieving) drugs. [22, 23] 2.

DISCUSSION

2.1. Machine Learning (ML) Assisted Drug Design

In the process of drug discovery, ML methods have mostly been used in all the above-mentioned computational approaches and other related fields over the past few decades (Appendix A, Figure 1). With the accumulation of “big” data on the molecular structures and their properties and activities, ML methods have developed into deep learning (DL) approaches to deal with important biological properties from large compound databases. ML and DL have been used in various applications in this context such as prediction of drug–protein interactions, discovery of drug efficacy, and ensuring the safety biomarkers (Appendix A, Figure 2). [24 28].

2.2. Machine Learning in Modern Anti-Inflammatory Drug Design and Discovery

Generally, ML-based drug discovery includes five steps: 1) target selection, 2) data collection, 3) data description, 4) model building, and 5) model evaluation. In the first step an appropriate target (usually a protein) is chosen which activity in inflammation process is already proved. If the study includes a structure-based approach, the small molecule libraries are searched for all potential structures that could inhibit or activate the target. If the ligand-based approach is followed, the similar elements in the structure of the small molecules that can inhibit or activate the target will be used as data. Afterwards, the most important step is to collect high-quality experimental data, which are obtained from specialized cheminformatic databases. Usually, two datasets including a positive dataset and a negative dataset are collected and a classification model is trained on this data. To characterize chemical structures of data as numeric features, two common description methods are utilized (molecular descriptors and molecular fingerprints). There are multiple online tools to calculate descriptors (e.g., PaDel-Descriptor [29] and E-Dragon [30]). The calculated descriptors could be mathematically further manipulated (e.g., Pearson correlation analysis) to improve the model [31]. After constructing the predictive models, performance metrics, including accuracy (Q), and Matthew’s correlation coefficient (MCC), and area under the receiver operating characteristic (ROC) curve (AUC) are used to evaluate the performance of the models. [32]

Novel NSAID Discovery

 In many studies the focus has been put on finding new molecular candidates, which inhibit COX but show fewer side effects than the classical NSAIDs. QSAR is one of the approaches that has been mostly used to investigate these structures. For example, a study published in 2022 involves the application of chemical descriptors and ML (22 ML classification models) to predict the bioactivity classes of molecules for COX-2 inhibition using real multidimensional COX-2 inhibitors (PubChem Fingerprints) obtained from a curated database (ChEMBL). Among all the tested algorithms, extreme gradient boosting classifier (XGB Classifier) was found to be the best performer (Q of 0.9484 and MCC of 0.8741). [33] In some other studies a combination of approaches has been used to repurpose existing drugs for anti-inflammatory effects. For example, the interaction sites between drugs and targets were determined by molecular docking. 67 drug-target interaction descriptors were calculated to describe the drug-target interactions, and 22 important descriptors were screened for drug classification by support vector machine (SVM), light gradient boosting method (LightGBM), and artificial neural network (ANN) (Q of 0.9329, 0.9268, and 0.9451, and MCC of 0.852, 0.840, and 0.882, respectively). Number of atom pairs, force field, and hydrophobic interactions were the key features for classifying anti-inflammatory drugs. [34]

2.2.2. Ant-inflammatory peptides (AIPs)

Although information about peptides are still scattered across the literature, several peptide-based databases exist that offer experimentally validated data [32] and are used in building predictive and design models of peptides drugs. Although these databases focus on specific type of peptides such as antimicrobial peptides and are not very comprehensive and most of them only cover the positive result of peptide activity, many of them play a crucial role in AIP discovery (e.g., APD [35], Peptaibol [36], CyBase [37], DADP [38], Hemolytik [39], AVPdb [40], DBAASP [41], SATPbd [42]). ML methods such as SVM, random forests (RF), and ANN have been widely applied to develop predictive models for peptide activity. [32] In 2017, Gupta et al. [43] developed predictive models to discriminate the AIPs from non-AIPs by utilizing SVM and RF methods (Q and MCC value of SVM model on validation set 0.72 and 0.45). The web server IL-10pred was constructed for the prediction of potential interleukin-10 (IL-10) inducing peptides, [44] which are one of the important cytokines, that suppress inflammatory responses, alleviate autoimmune pathologies [45] and extend graft survival [46]. This server was developed based on RF algorithm by utilizing dipeptide composition, which displayed the highest Q and MCC value of 0.8224 and 0.59. AIP-Pred [47] is a web server, which focuses on identifying potential AIP candidates using extremely randomized tree (ERT), SVM, k-nearest neighbor (kNN) and RF approaches combined with the selected optimal features to build classifiers. The AUC value of the best classifier achieved 0.814 on an independent dataset. Apart from the databases and online predictive tools, there are many ML-based in silico studies for predicting peptide activity that are not integrated to web servers. For example, AntiFlamPred uses a deep neural network (DNN) approach to predict the anti-inflammatory response of peptides with an AUC of 0.919 and MCC of 0.735. [48]

2.2.3. Anti-inflammatory plant-based compounds and natural products

 Usually, compounds obtained from natural sources have little or no side effects, thus searching for new lead compounds among traditionally used anti-inflammatory medications is a common strategy. Therefore, high throughput in silico ML screening methods to identify potential compounds from natural sources is a growing field in drug discovery efforts. In addition to COX, lipoxygenase pathway is important in inflammatory processes; therefore, natural inhibitors of both pathways would act as high efficacy and low side effect anti-inflammatory medications. In one of the early studies with this metabolomic approach, leaf extracts of Asteraceae species with in vitro dual inhibition of cyclooxygenase-1 and 5-lipoxygenase were analyzed by high performance liquid chromatography and high-resolution mass spectrometry analysis using ML algorithms. As a result, a robust ANN model for predicting the anti-inflammatory activity of natural compounds was obtained (Q of 0.81). [49-51] This approach has been used by multiple other studies to inhibit the activity of inflammatory biomarkers. [52-55] In another study, three ML approaches, including SVM, RF and GBM have been applied to develop the classification model. The model was generated using the cyclooxygenase-2 (COX-2) inhibitors reported in the ChEMBL database (Q of SVM, RF and GBM was found to be 0.7540, 0.7479, and 0.7460, respectively). Application of the model on NuBBE database, a repository of natural compounds, led the researchers to identify a natural compound, enhydrin, possessing analgesic and anti-inflammatory activities. [56] Recently, a comprehensive online web platform (InflamNat) has been developed for anti-inflammatory natural product research. InflamNat is a database containing the physicochemical properties, cellular anti-inflammatory bioactivities, and molecular targets of 1351 natural products that tested on their anti-inflammatory activities. It also provides two ML-based predictive tools specifically designed for natural products that (1) predict the anti-inflammatory activity of natural products and (2) predict the compound-target relationship for compounds and targets collected in the database but lacking existing relationship data. A novel multi-tokenization transformer model was used as the sequential encoder for both predictive tools (AUC of 0.842 and 0.872 in anti-inflammatory activity prediction and compound-target interactions, respectively). [57]

CONCLUSION

The process of designing and developing new drugs including anti-inflammatory drugs is extremely expensive and cumbersome. However, currently available anti-inflammatory medications (corticosteroids and NSAIDs) have severe side effects. Additionally, the new advancements in computational biology, cheminformatics, and bioinformatics, unveil the molecular mechanisms behind inflammation with a speed we have never experienced in science before. Scientists have used this opportunity to apply ML assisted methods towards the anti-inflammatory drug design and discovery (structure-based and ligand-based approaches). These efforts have been taken in three directions: 1) ML assisted discovery of novel NSAIDs with less side effects, 2) ML assisted discovery of AIPs with high specificity, 3) ML assisted discovery of anti-inflammatory natural products with no side effects. Generally, the best results are the product of merging some of the classical approaches in drug design and discovery (e.g., docking and QSAR) with the state-of-the-art ML and DL methods.

REFERENCE

1. Lin, X., X. Li, and X. Lin, A review on applications of computational methods in drug screening and design. Molecules, 2020. 25(6): p. 1375.
2. Willems, H., S. De Cesco, and F. Svensson, Computational Chemistry on a Budget: Supporting Drug Discovery with Limited Resources: Miniperspective. Journal of Medicinal Chemistry, 2020. 63(18): p. 10158 10169.
3. Decherchi, S. and A. Cavalli, Thermodynamics and kinetics of drug-target binding by molecular simulation. Chemical Reviews, 2020. 120(23): p. 12788-12833.
4. Nunes-Alves, A., D.B. Kokh, and R.C. Wade, Recent progress in molecular simulation methods for drug binding kinetics. Current Opinion in Structural Biology, 2020. 64: p. 126-133.
5. Maia, E.H.B., et al., Structure-based virtual screening: from classical to artificial intelligence. Frontiers in chemistry, 2020. 8: p. 343.
6. Garcia-Hernandez, C., A. Fernandez, and F. Serratosa, Ligand-based virtual screening using graph edit distance as molecular similarity measure. Journal of chemical information and modeling, 2019. 59(4): p. 1410-1421.
7. da Silva Rocha, S.F., et al., Virtual screening techniques in drug discovery: review and recent applications. Current topics in medicinal chemistry, 2019. 19(19): p. 1751-1767.
8. Sethi, A., et al., Molecular docking in modern drug discovery: Principles and recent applications. Drug discovery and development-new advances, 2019. 2: p. 1 21.
9. Pinzi, L. and G. Rastelli, Molecular docking: shifting paradigms in drug discovery. International journal of molecular sciences, 2019. 20(18): p. 4331.
10. Sun, H., Pharmacophore-based virtual screening. Current medicinal chemistry, 2008. 15(10): p. 1018 1024.
11. Caporuscio, F. and A. Tafi, Pharmacophore modelling: a forty-year-old approach and its modern synergies. Current medicinal chemistry, 2011. 18(17): p. 2543 2553.
12. Muratov, E.N., et al., QSAR without borders. Chemical Society Reviews, 2020. 49(11): p. 3525-3564.
13. Neves, B.J., et al., QSAR-based virtual screening: advances and applications in drug discovery. Frontiers in pharmacology, 2018. 9: p. 1275.
14. Hartenfeller, M. and G. Schneider, De novo drug design. Chemoinformatics and computational chemical biology, 2010: p. 299-323.
15. Nathan, C. and A. Ding, Nonresolving inflammation. Cell, 2010. 140(6): p. 871-882.
16. Okin, D. and R. Medzhitov, Evolution of inflammatory diseases. Current Biology, 2012. 22(17): p. R733-R740.
17. Kapugi, M. and K. Cunningham, Corticosteroids. Orthopaedic Nursing, 2019. 38(5): p. 336-339.
18. Brune, K. and P. Patrignani, New insights into the use of currently available non-steroidal anti-inflammatory drugs. Journal of pain research, 2015. 8: p. 105.
19. Buchman, A.L., Side effects of corticosteroid therapy. Journal of clinical gastroenterology, 2001. 33(4): p. 289 294.
20. Scheiman, J.M., NSAID-induced gastrointestinal injury. Journal of clinical gastroenterology, 2016. 50(1): p. 5-10.
21. Dara, S., et al., Machine learning in drug discovery: a review. Artificial Intelligence Review, 2022. 55(3): p. 1947-1999.
22. Smith, J.S., A.E. Roitberg, and O. Isayev, transforming computational drug discovery with machine learning and AI. 2018, ACS Publications. p. 1065-1069.
23. Bender, A. and I. Cortés-Ciriano, Artificial intelligence in drug discovery: what is realistic, what are illusions? Part 1: ways to make an impact, and why we are not there yet. Drug discovery today, 2021. 26(2): p. 511-524.
24. Bender, A. and I. Cortes-Ciriano, Artificial intelligence in drug discovery: what is realistic, what are illusions? Part 2: a discussion of chemical and biological data. Drug Discovery Today, 2021. 26(4): p. 1040-1052.
25. Ghasemi, F., et al., Neural network and deep-learning algorithms used in QSAR studies: merits and drawbacks. Drug discovery today, 2018. 23(10): p. 1784-1790.
26. Yap, C.W., PaDEL?descriptor: An open source software to calculate molecular descriptors and fingerprints. Journal of computational chemistry, 2011. 32(7): p. 1466-1474. 30. Tetko, I.V., et al., Virtual computational chemistry laboratory–design and description. Journal of computer aided molecular design, 2005. 19(6): p. 453-463.
27. Fang, J., et al., Discovery of neuroprotective compounds by machine learning approaches. RSC advances, 2016. 6(12): p. 9857-9871.
28. Wu, Q., et al., Recent progress in machine learning based prediction of peptide activity for drug discovery. Current topics in medicinal chemistry, 2019. 19(1): p. 4 16.
29. Dibia, K.T., et al., Exploration of the quantitative Structure-Activity relationships for predicting Cyclooxygenase-2 inhibition bioactivity by Machine learning approaches. Results in Chemistry, 2022. 4: p. 100272.
30. Huang, S. and Y. Ding, Identification of Anticancer and Anti-inflammatory Drugs from Drugtarget Interaction Descriptors by Machine Learning. Letters in Drug Design & Discovery, 2022. 19(9): p. 800-810.
31. Peptaibol. http://peptaibol.cryst.bbk.ac.uk/home.shtml
32. CyBase. https://www.cybase.org.au/ 38. DADP. http://split4.pmfst.hr/dadp/
33. Hemolytik. https://academic.oup.com/nar/article/42/D1/D444/10416 83
34. AVPdb. https://academic.oup.com/nar/article/42/D1/D1147/1058 993
35. DBAASP. https://dbaasp.org/
36. SATPdb. https://academic.oup.com/nar/article/44/D1/D1119/2502 618
37. Gupta, S., et al., Prediction of anti-inflammatory proteins/peptides: an insilico approach. Journal of translational medicine, 2017. 15(1): p. 1-11.
38. Nagpal, G., et al., Computer-aided designing of immunosuppressive peptides based on IL-10 inducing potential. Scientific reports, 2017. 7(1): p. 1-10.
39. Hawrylowicz, C. and A. O'garra, Potential role of interleukin-10-secreting regulatory T cells in allergy and asthma. Nature Reviews Immunology, 2005. 5(4): p. 271-283.
40. Bromberg, J.S., IL-10 immunosuppression in transplantation. Current opinion in immunology, 1995. 7(5): p. 639-643.
41. Manavalan, B., et al., AIPpred: sequence-based prediction of anti-inflammatory peptides using random forest. Frontiers in pharmacology, 2018. 9: p. 276.
42. Alotaibi, F., M. Attique, and Y.D. Khan, AntiFlamPred: an anti-inflammatory peptide predictor for drug selection strategies. Comput Mater Contin, 2021. 69: p. 1039-1055.
43. Chagas-Paula, D., et al., Discovery of plant anti-inflammatory biomarkers by machine learning algorithms and metabolomic studies. Planta Medica, 2013. 79(13): p. SL27.
44. Chagas-Paula, D.A., et al., Prediction of anti-inflammatory plants and discovery of their biomarkers by machine learning algorithms and metabolomic studies. Planta Medica, 2015. 81(06): p. 450-458.
45. Chagas-Paula, D.A., et al., A metabolomic approach to target compounds from the Asteraceae family for dual COX and LOX inhibition. Metabolites, 2015. 5(3): p. 404-430. 52.Usman, A.G., et al., The effect of ethanolic leaves extract of Hymenodictyon floribundun on inflammatory biomarkers: a data-driven approach. Bulletin of the National Research Centre, 2021. 45(1): p. 1-12.
46. Nicacio, K.d.J., et al., Anti?Inflammatory Markers of Hops Cultivars (Humulus lupulus L.) Evaluated by Untargeted Metabolomics Strategy. Chemistry & Biodiversity, 2022. 19(4): p. e202100966.
47. Ferreira, M.S., et al., Systematic Review of Anti-inflammatory Agents from Aspergillus Species. Revista Brasileira de Farmacognosia, 2021: p. 1-12.
48. Domingos, O.d.S., et al., Anti-inflammatory derivatives with dual mechanism of action from the metabolomic screening of Poincianella pluviosa. Molecules, 2019. 24(23): p. 4375.
49. Khan, M.F., R.B. Rashid, and M.A. Rashid, Identification of Natural Compounds with Analgesic and Anti-inflammatory Properties Using Machine Learning and Molecular Docking Studies. Letters in Drug Design & Discovery, 2022. 19(3): p. 256-262.
50. Zhang, R., et al., InflamNat: web-based database and predictor of anti-inflammatory natural products. Journal of Cheminformatics, 2022. 14(1): p. 1-11.