How In silico Innovation is Transforming Cosmetic Ingredient Discovery and Development

Advances in the cosmetics industry are often held as trade secrets by individual companies. Like the nutraceutical and pharmaceutical industries, the cosmetics industry uses fine chemicals – but historically, it has done so, in part, for a different motivation: such additives are often employed not for functional benefits (as formulated), but rather for marketing purposes or to distinguish the brand.

With growing competition and advances in innovation, however, additives with true functionality could help the industry keep up with consumer expectations – which in today’s market, requires the production of advanced, scientifically proven actives that not only perform better than the benchmarks, but also set new industry standards.¹

The range of potential ingredients and actives for cosmetic applications is vast yet insufficiently investigated. Whereas more than 500,000 natural compounds and therapeutic plants are known and used in different businesses,³ the International Nomenclature of Cosmetic Ingredients (INCI) list contains only 16,000 approved ingredients.⁴ This gap underscores the immense opportunity to reveal novel ingredients and actives with benefits that have not been considered for cosmetic use.

Developing new and efficacious actives is critical for the industry² but identifying and incorporating modern ingredients such as vitamins, plant extracts and bioactive compounds into cosmetics has been resource intensive. The traditional discovery process takes considerable time and money, and can limit the capacity to identify new ideas. This strategy often involves large-scale random screening, which leads to lower success rates and higher development costs because many compounds are examined with no prior knowledge.

As an alternative, promising candidate molecules can be manually synthesized and optimized, but this also involves trial-and-error experimentation, which can be both costly and inefficient – and to complete the product development process, extensive in vitro and in vivo testing are needed to assess the ingredient’s unknown stability, safety and efficacy.² While these tests are necessary whether a material’s properties are known or unknown, the initial screening can be made less burdensome.

Modern computational techniques are used throughout the pharmaceutical industry to rapidly identify likely candidates, and this approach can be applied to the cosmetics industry. Accelerating discovery with computational methods and artificial intelligence (AI) can methodically explore and prioritize promising candidates, opening new potential for effective ingredients that meet customer expectations for viability, naturality and sustainability. Machine learning (ML) algorithms, coupled with modeling techniques based on material science, can address complex issues related to skin and hair well-being, ensuring and enhancing skin and hair health and appearance.

This article takes a closer look at the potential applications of in silico (computer calculated) approaches to optimize cosmetic ingredient discovery — from virtual screening and quantitative structure activity relationship analyses (QSAR), to repurposing existing molecules and predicting their interactions with skin. A full computational pipeline, along with a case study of a psoriasis peptide in development, is also presented to illustrate how these technologies can unlock solutions to the consumer demand for high-performance, scientifically backed formulations.

Accelerating Discovery with Computational In silico, ML and AI Methods

Similar to the drug discovery process, the cosmetic ingredient development process can begin by identifying important target proteins for hair and skin care. It should also determine so-called excellence compounds – those that exhibit desirable physicochemical properties in cosmetic products.⁵ By leveraging AI-driven calculations and atomic modeling strategies, described next, it is possible to analyze complex atomic structures; anticipate their interaction with human hair, nail and skin; and assess their safety and efficacy profiles. This data-driven technique has resulted in not only more productivity, but also more scientific evidence for innovative, high-performance cosmetic products.

Computational Chemistry

Computational chemistry has risen as a crucial component in the rational design and optimization of active ingredients for cosmetic and dermaceutical applications. Both physics-based simulations and data-driven approaches contribute to distinguishing compounds with upgraded target affinity, favorable pharmacokinetic behavior and improved metabolic profiles. These techniques not only help in refining existing molecules, but also support the de novo development of new compounds by methodically assembling molecular structures with desired functional and physicochemical properties.

The efficiency of screening compound libraries for useful activity has altogether improved with recent progress in molecular modeling and cheminformatics. These tools are especially useful in assessing underexplored chemical platforms, including derivatives of natural products and synthetic analogs, for their potential biological effects. The ability to analyze structure–activity relationships computationally provides more profound insight into the variables influencing efficacy and safety, thus directing the plan of fundamentally optimized candidates.⁵

Moreover, virtual screening methods integrated early in the discovery pipeline allow for the quick evaluation of molecular traits such as solubility, stability, bioavailability and toxicity. This in silico filtering makes a difference by prioritizing compounds with the most promising profiles. ML models help to determine bioactivity and rule out inactive or low-potential structures (see Figure 1). Together, these computational approaches contribute to a more focused and resource-efficient pipeline.

Computational chemistry can be useful in:

• Predictive modeling for small molecule and peptide-based active ingredients
• Physics-based and machine learning models for skin-targeted actives
• Accelerated discovery of target-specific molecules using AI tools
• Integration of cheminformatics for functional ingredient identification
• Optimization of lead compounds by increasing their binding affinity and improving their pharmacokinetic and metabolic properties
• Discovery of entirely new compounds

Predictive Modeling Using ML and AI discovery

The use of ML allows for the prediction and targeting of bioactive compounds that target specific skin-related proteins from molecular descriptors and experimental activity data. This approach is known as anticipatory modeling. Typically, the workflow comprises a computational pipeline that includes data preprocessing for features and functional training to models using classification or regression algorithms, as well as cross-validation for performance evaluation. Time and cost associated with experimental screening can be significantly decreased by prioritizing compounds that have high predicted efficacy, as well as desirable safety profiles, using these models.

Formulation science expertise is crucial in translating active compounds into stable, effective end products, although ingredient development is increasingly important. None of the new in silico methods changes the overall processes involved, they just accelerate it – the goal remains to develop unique, targeted formulations that can be made at large scale. Additionally, practical expertise in optimizing formulation parameters, such as rheology, stability and sensory attributes, ensures that computationally selected actives are effectively integrated into ready-for-market formulations.⁶

To maximize chances for success, a multi-step computational pipeline, adapted from established datasets and cross-validated, can be incorporated. Following is a closer look.⁷

Step 1 – data preprocessing. The available dataset of compounds necessary for the modeling should be filtered to retain only the simplified molecular input line entry system (SMILES) strings – the compact, ASCII-based line notations used to represent 2D or 3D chemical molecular structures for computer processing; physico-chemical properties; and relevant pIC₅₀ values – i.e., the negative logs of the molar IC₅₀ values representing compound potency.

All redundant columns should be removed, and all duplicate and null values should be discarded. An “activity” column is created, where compounds will be classified based on a threshold pIC₅₀ value — e.g., compounds having a pIC₅₀ of less than 10 μM are labeled as active, while those greater are deemed inactive. This threshold must be set on a case-by-case basis.

Step 2 – incorporating decoys: To achieve enhanced robustness of the model, decoys sourced from the Directory of Useful Decoys, Extended (DUDE-Z) — for tyrosinase, in the case of psoriasis — are introduced into the dataset. This balancing step helps to maintain an appropriate active-to-inactive ratio (at least 1:9) for robust binary classification in the ML models. This must be done with caution, as many compounds will not have been evaluated against the target and may in fact be actives, although the probability of this is quite low if the activity threshold is set at a low value.

Step 3 – descriptor generation and selection: Molecular descriptors are generated by converting SMILES into fingerprints (e.g., Morgan, RDK, MACCS) that are numerical representations of chemical structures.²⁷ Coupled with these descriptors are certain molecular properties (molecular weight, logP, hydrogen bond counts, etc.) that provide a full molecular representation. This is necessary for successful model training. 

It is difficult a priori to identify which molecular properties might be useful for the model, so it is recommended to include as much available and reliable information as possible. Caution should be taken when using data calculated via other methods, as these are approximate only and could propagate error in the model.

Step 4 – model training and selection: A few probabilistic models have been extensively evaluated in the scientific and technical literature. Examples include: Random Forest, XGBoost, Support Vector Machine (SVM), k-Nearest Neighbors and Logistic Regression. Each of these can perform best for a given use case depending on the availability of data and the need for streamlining the process.

Although an experienced data scientist can prioritize certain approaches, at this point, determining the best model remains empirical; for curious technically minded readers, some excellent comparative reviews are available by searching the literature. In the specific example given here, the XGBoost classifier with Morgan fingerprints performed well and balanced computational cost with accuracy.

Step 5 – hyperparameter tuning and evaluation: Hyperparameters are tuned to maximize the weighted F1-score, a machine learning metric, as the objective function. The evaluation metrics, including the F1 scores, assert the effectiveness of the model in discriminating active from inactive compounds.

Step 6 – computational setup: All of the previous steps can be performed, if slowly, using any modern computer but to train the model and to use it to generate data, access to a computer cluster is necessary. These are available for hire from cloud services. In the present case, a cluster with 32 CPUs and 24 processes running in parallel was sufficient for data management purposes. This low intensity system was only possible by eliminating extraneous descriptors, thus enabling efficient model building.

Step 7 – virtual screening and predictions: The trained model can then be applied for high-throughput virtual screening, classifying compounds based on a probability threshold for activity. This stringent criterion yielded a shortlist of potential candidates for further experimental validation (see Figure 2). Available libraries of compounds, properly parameterized for most uses, are available from many contract research organizations that offer small molecules to make it easy for the customer to identify molecules to purchase.

Quantitative Structure-Activity Relationship (QSAR) Method

QSARs with predictive power are the next stage of developing predictive models for the synthesis of new molecules with desired physicochemical and biological properties. These models allow for the design of target-specific actives, with interests both from the scientific world and the commercial markets simultaneously ensuring relevance to both ends and consumer appeal.

Data-centric modeling thus brings speed to the entire discovery pipeline, ensuring experts work on the most likely solution to succeed, thus cutting costs and time-to-market. This ensures the faster output of differentiated products for fulfilling unmet needs in skin and hair care. The combination of computational design with predictive modeling and relevant experimental validation is thereby a means to open new avenues in active ingredient innovation and envision the next generation of truly high-performance cosmetic solutions.⁸

Practical Applications of ML-Driven In silico Design: Skin Permeation

Skin permeation: An important application of computational methods for cosmeceutical research is predicting active ingredient interactions and skin permeation. Molecular dynamics simulations (MD) can provide highly detailed models of active permeation through the skin lipid layers and barrier structures, to understand how deeply they may permeate and how potentially effective they could be. 

In a complementary manner, QSAR modeling and pharmacophore analysis – to identify the 3D spatial arrangement of essential chemical features – assist scientists in modifying the active ingredient structures for better absorption, better safety profile and specific biological effect(s).⁹

Repurposing known ingredients: Computational strategies also can facilitate the repurposing of existing cosmetic ingredients by identifying new functions. AI algorithms and ML paradigms can be used to scan chemical databases for subtle chemical likenesses and patterns that could relate compounds with unforeseen benefits, such as anti-inflammatory, antioxidant or skin brightening. This set of actions therefore broadens the functionality of ingredients previously considered to be well-known.¹⁰

Fast-tracking research: AI-driven data analysis can also act as a catalyst for research activity by working through massive volumes of scientific and experimental information. Advanced literature mining tools scour research articles, patents and experimental datasets in the hopes of finding promising actives and emerging trends – although it is worth noting, caution must be taken when integrating this data into any model. One should always confirm promising results through a human reading of the sources. 

Still, predictive analyses are capable of forecasting ingredient performance, efficacy and safety before the product hits the lab bench, shaving precious time off its time-to-market. Further, computational platforms that integrate multi-omics data including genomics, proteomics and metabolomics will be a powerful predictive tool.

Leveraging Computational Workflows to Accelerate Discovery

From the bigger picture view, in silico drug discovery pipelines integrate computational methods to streamline every step in the process (see Figure 3); from target identification (hit finding) to compound discovery and optimization. 

Briefly, the initial step involves screening vast chemical libraries — e.g., up to 80 billion compounds from the REAL Enamine database, or more than 10 trillion in xReal; other sources report even higher numbers — whose molecular datasets must undergo data preprocessing and filtering to focus on structure selection and mutation to find protein candidates with promising drug-like properties. 

Following this protein preparation, initial molecular dynamics (MD) simulations are carried out using a portal validated to separate actives from decoys. SiteMap analysis identifies potential binding pockets, and molecular docking predicts ligand interactions. A second MD simulation then refines the binding state, enabling accurate binding-free energy calculations using Molecular Mechanics/Generalized Born Surface Area (MM-GBSA) and Free Energy Perturbation (FEP) methods.

Subsequent protein-ligand interaction analysis and experimental techniques such as Isothermal Titration Calorimetry (ITC) and Surface Plasmon Resonance (SPR) are used to validate binding affinity and mechanism, completing the cycle from virtual prediction to laboratory confirmation. 

Thus, the integrated workflow fast-tracks hit-finding, makes hits better, and facilitates the design of lead molecules for further experimental verification (see Figure 3).¹¹ With such an integrated approach, hits undergo extensive validation and provide a framework for the rational selection of leads for experimental validation.¹³

Integrated discovery platforms can assist with identification and optimization processes, narrowing candidates to create receptor-targeted libraries. For example, the authors have trimmed their company's library of 4.5 billion compounds to a curated set of 50 million drug-like candidates, then filtered these further to ~3.2 million synthetically feasible small peptides for further screening¹² — using a platform capable of managing up to 506 terabytes of data storage capacity.

Such tools were applied in an ongoing study exploring peptides to treat psoriasis.

Dermaceutical Peptides and Psoriasis

Peptides are a significant growth area. For example, they currently constitute about 11% of pharmaceutical industry sales – almost US $24 billion in 2017 – and this market is expected to grow to US $43 billion by 2024.¹⁴ For skin care, the development of novel topically applied peptides that can penetrate the skin and help support the extracellular matrix is of interest, particularly to tighten the skin and remove wrinkles.

A major subclass of peptides is collagen pentapeptides.^15-17 These short sequences, typically with the sequence KTTKS, have shown extensive anti-aging benefits. These minimal fragments (residues 212-216) of pro-collagen are sufficient to increase collagen production in fibroblasts.^{18, 19}

These materials have been studied both for their physical behavior and biological activity.^18-21 Pentapeptides can inhibit collagenases and MMP-9 activity, preventing extracellular matrix (ECM) breakdown.^{22, 23} However, the dominant mechanism of action of KTTKS has yet to be conclusively established.²⁴

Furthermore, peptides can be designed to minimize side effects since they can be far more selective inhibitors than small molecules, thanks to their increased size. Therefore, peptide interactions are often more specific.²⁵ Short peptides have been shown to meaningfully decrease the activity of Protein Kinase C, although they have not been sufficiently potent to be clinically relevant.²⁶

Generally speaking, the problem with peptide products intended for use in vivo is stability. Peptides are readily degraded by peptidases and this means bioavailability can often be limited. The additional challenge with any topical formulation, as is implied for the cosmetic industry, is skin permeation. The final challenge in this chemical space is intellectual property and the development of targeted molecules.

The latter challenge can be addressed by computationally screening cyclic and polycyclic peptides for their interaction with PKC. The “hit” peptides revealed through this screening then serve as starting points for further elaboration by integrating unnatural residues to improve affinity for specific receptors and to increase stability. These can also mimic post-translational modifications to improve selective skin uptake, like through palmitoylation.

Case Study: Targeting Proteins Implicated in Psoriasis

Psoriasis is an autoimmune inflammatory skin disease that affects 0.5–3.0% of the world’s population.²⁸ It is caused by a complex interplay of the innate and adaptive immune systems with a wide array of genetic and environmental factors. Environmental triggers such as stress, injury and drugs play a role in starting the self-propelled cycle of inflammation, which culminates in hyper-proliferation due to the activation of the innate immune system.

Topical corticosteroids are useful and often the first line of treatment for limited or small skin areas affected by psoriasis. For example, due to its effects on calcium metabolism, calcipotriene ointment is useful in treating psoriasis.²⁹

On the other hand, oral medications include methotrexate, acitretin, cyclosporine, apremilast and others; in fact, drugs have been developed against fifteen selected and validated anti-psoriatic targets. However, the current drugs prescribed for psoriasis management have limitations, ranging from the lack of potency (topical agents) and high toxicity (anticancer agents), to high cost and relatively large molecular size.

Studies report that the interaction between T-cells and keratinocytes gives rise to a cytokine “soup” dominated by Th1- type and Th17-type cytokines such as interleukin- (IL-) 12, IL-17, interferon- (IFN-) γ and tumor necrosis factor (TNF). In addition, keratinocytes stimulated with IL-20 upregulate a variety of inflammatory genes, including monocyte chemotactic protein-1 (MCP-1) and myeloid-related protein-14 (MRP-14).³⁰ TNF is currently considered a major target in psoriasis pathogenesis because much higher levels of TNF are found in lesional skin than in normal skin.³¹

Antimicrobial peptides and proteins (AMPs) also play an important role in the pathogenesis of psoriasis.³² As a result, etanercept, a blocker of the pro-inflammatory cytokine tumor necrosis factor-α (TNF-α), is effective in the treatment of psoriasis.³³ It has been reported that the peptide LL37, or CAMP, drives autoimmunity in psoriasis and may also play a role in other autoimmune disorders, since the peptide is heavily expressed in maladies such as inflammatory bowel disease and rheumatoid arthritis.³⁴ 

Pharmacological studies on peptide T and its analogs also point to the potential applicability of such peptides as therapeutic agents for the treatment of neuropsychometric symptoms in AIDS patients and psoriasis.^{35, 36}

Therapeutic peptides are a promising and a novel approach to treat many diseases, including cancer. Such peptides have several advantages over therapeutic proteins or antibodies, since they have high activity, affinity, target specificity, selectivity, biological and chemical diversity, low toxicity, minimal drug-drug interactions and the ability to penetrate cell membranes. 

An added benefit of using peptides as a treatment is that they do not accumulate in specific organs (e.g., kidney or liver), which can help to minimize any toxic side effects.³⁷ Further, peptides are small, relatively easy to synthesize and modify, and are less immunogenic than recombinant antibodies or proteins. Owing to these properties, therapeutic peptides show great potential in the treatment of many diseases.

The use of AI, ML and computational tools has become a common practice in drug research. Besides time and cost savings, advances in technology and computing power have significantly improved the accuracy of theoretical results, ensuring a close match between computational predictions and experimental outcomes.

In current work, these methodologies are being used to design and screen new peptides and cyclic peptides – as well as to repurpose known therapeutic peptides – against fifteen selected and validated anti-psoriatic drug targets. Molecular docking and MD simulation are being used to determine strong binders for disease treatment (see Figure 4).

Analysis will then be carried out to identify drugs with better docking scores than co-crystallized ligands across the entirety of protein targets. The goal is to identify multi-targeted peptides that have the ability to bind well to all targets. The selected peptides will then be synthesized to validate the anti-psoriatic property of selected peptides in a biological assay.

Conclusion

In the cosmetics and personal care industry, next-generation innovation must surpass traditional methods of trial-and-error prospecting. As this paper intends to show, advanced computational techniques including machine learning, QSAR modeling and virtual screening can facilitate the identification of active ingredients with more emphasis and efficiency. These processes can not only achieve greater accuracy in predicting efficacy, safety and skin compatibility, they can also cut down on R&D costs and shorten the manufacturing period.

Psoriasis provides a case study demonstrating how in silico design is facilitating the identification of novel dermaceutical peptides with good potential efficacy. Thus, the use of computational intelligence in formulation science will enable the creation of a different category of personalized, scientifically grounded cosmetic treatments that meet ever-changing market needs and pinpoint innovation.

References

1. Zappelli, C., Barbulova, A., Apone, F. and Colucci, G. (2016). Effective active ingredients obtained through biotechnology. Cosmetics, 3.
2. Nohynek, G.J., Antignac, E., Re, T. and Toutain, H. (2010). Safety assessment of personal care products/cosmetics and their ingredients. Toxicology and Applied Pharmacology 243, 239-259.
3. Ntie-Kang, F. and Svozil, D. (2020). An enumeration of natural products from microbial, marine and terrestrial sources. Physical Sciences Reviews, 5(8); https://doi.org/doi:10.1515/psr-2018-0121
4. Davis, C.C. and Choisy, P. (2024). Medicinal plants meet modern biodiversity science. Current Biology 34, R158-R173; https://doi.org/https://doi.org/10.1016/j.cub.2023.12.038
5. El Bouamri, L., Bouachrine, M. and Chtita, S. (2025). Computational studies in dermo-cosmetics: In silico discovery of therapeutic agents targeting a variety of proteins for skin diseases. Curr Top Med Chem 25, 657-688; https://doi.org/10.2174/0115680266337405240926114604
6. Di Guardo, A., et al. (2025). Artificial intelligence in cosmetic formulation: Predictive modeling for safety, tolerability and regulatory perspectives. Cosmetics 12.
7. Goussard, V., et al. (2020). A new machine-learning tool for fast estimation of liquid viscosity. Application to cosmetic oils. J Chemical Information and Modeling 60, 2012-2023; https://doi.org/10.1021/acs.jcim.0c00083
8. Kim, J.Y., Kim, M.K., Kim, K.-B., Kim, H.S. and Lee, B.-M. (2019). Quantitative structure–activity and quantitative structure–property relationship approaches as alternative skin sensitization risk assessment methods. J Toxicology and Environmental Health, Part A 82, 447-472; https://doi.org/10.1080/15287394.2019.1616437
9. Matthew, B. (2025) Application of machine learning to predict enhancer skin interaction profiles. Available at https://www.researchgate.net/publication/393789798
10. Grabenhofer, R. (2024). Digital tech in beauty: Ingredient discovery, early trend spotting, personalization, digital collabs, molecular design and more. Cosm & Toil. Available at https://www.cosmeticsandtoiletries.com/research/consumers-market/article/22922396
11. Agamah, F.E., et al. (2020). Computational/in silico methods in drug target and lead prediction. Brief Bioinform 21, 1663-1675; https://doi.org/10.1093/bib/bbz103
12. Guo, S., Zhao, T., Yun, Y. and Xie, X. (2022). Recent progress in assays for GPCR drug discovery. Amer J Physiology-Cell Physiology 323, C583-C594; https://doi.org/10.1152/ajpcell.00464.2021
13. Moshawih, S., et al. (2024). Consensus holistic virtual screening for drug discovery: A novel machine learning model approach. J Cheminform 16, 62; https://doi.org/10.1186/s13321-024-00855-8
14. Zion Market Research. (2018, Oct 18). Peptide therapeutics market by route of administration (parenteral, oral and others), by type (innovative and generic), by type of manufacturers (in-house and outsourced), by type of molecule (vasopressin, somatostatin, calcitonin, immunopeptide, natriuretic and others), by technology (liquid phase, solid phase and hybrid phase) and by application (cancer, cardiovascular disease, metabolic disease, respiratory disease, infectious disease and others): Global industry perspective, comprehensive analysis and forecast, 2017 – 2024. Available at https://www.zionmarketresearch.com/news/peptide-therapeutics-market
15. Katayama, K., Armendariz-Borunda, J., Raghow, R., Kang, A.H. and Seyer, J.M. (1993). A pentapeptide from type I procollagen promotes extracellular matrix production. J Biological Chem, 268, 9941-9944.
16. Clifford, T., et al. (2019). The effects of collagen peptides on muscle damage, inflammation and bone turnover following exercise: A randomized, controlled trial. Amino Acids 51, 691-704; https://doi.org/http://doi.org/10.1007/s00726-019-02706-5
17. De Luca, C., et al. (2016). Skin anti-aging and systemic redox effects of supplementation with marine collagen peptides and plant-derived antioxidants: A single-blind case-control clinical study. Oxidative Medicine and Cellular Longevity, 4389410-4389410 (2016); https://doi.org/10.1155/2016/4389410
18. Jones, R.R., Castelletto, V., Connon, C.J. and Hamley, I.W. (2013). Collagen stimulating effect of peptide amphiphile C16–KTTKS on human fibroblasts. Molecular Pharmaceutics, 10, 1063-1069 (2013); https://doi.org/10.1021/mp300549d
19. Castelletto, V., Hamley, I.W., Perez, J., Abezgauz, L. and Danino, D. (2010). Fibrillar superstructure from extended nanotapes formed by a collagen-stimulating peptide. Chemical Communications 46, 9185-9187; https://doi.org/10.1039/c0cc03793a
20. Palladino, P., Castelletto, V., Dehsorkhi, A., Stetsenko, D. and Hamley, I.W. (2012). Conformation and self-association of peptide amphiphiles based on the KTTKS collagen sequence. Langmuir 28, 12209-12215; https://doi.org/10.1021/la302123h
21. Castelletto, V., et al. (2013). Self-assembly of palmitoyl lipopeptides used in skin care products. Langmuir 29, 9149-9155; https://doi.org/http://doi.org/10.1021/la401771j
22. Wells, J.M., Gaggar, A. and Blalock, J.E. (2015). MMP generated matrikines. Matrix Biology 44-46, 122-129; https://doi.org/https://doi.org/10.1016/j.matbio.2015.01.016
23. Lupo, M.P. and Cole, A.L. (2007). Cosmeceutical peptides. Dermatologic Therapy 20, 343-349; https://doi.org/10.1111/j.1529-8019.2007.00148.x
24. Abu Samah, N.H. and Heard, C.M. (2011). Topically applied KTTKS: A review. Intl J Cos Sci, 33, 483-490; https://doi.org/10.1111/j.1468-2494.2011.00657.x
25. Meister, D., Taimoory, S.M. and Trant, J.F. (2018). Unnatural amino acids improve affinity and modulate immunogenicity: Developing peptides to treat MHC Type II autoimmune disorders. Pept Sci, 111, e24058; http://doi.org/10.1002/pep2.24058
26. Park, H.-Y., et al. (2004). Topical application of a Protein Kinase C inhibitor reduces skin and hair pigmentation. J Invest Dermatol, 122, 159-166; https://doi.org/https://doi.org/10.1046/j.0022-202X.2003.22134.x
27. Gao, Y., Mu, J., Liu, K. and Wang, M. (2025 Sep). Integrating molecular fingerprints with machine learning for accurate neurotoxicity prediction: An observational study. Advanced Technology in Neuroscience. Available at https://journals.lww.com/atn/fulltext/2025/09000/integrating_molecular_fingerprints_with_machine.2.aspx
28. Ayala-Fontánez, N., Soler, D.C and McCormick, T.S. (2016, Feb 22). Current knowledge on psoriasis and autoimmune diseases. Psoriasis. Available at https://pmc.ncbi.nlm.nih.gov/articles/PMC5683130/
29. Medline Plus. (Accessed 2026, Feb 10). Calcipotriene topical. National Library of Medicine. Available at https://medlineplus.gov/druginfo/meds/a608018.html
30. Park, D., Jeong, H.O., Byoung-Chul, K., Young, M.H. and Hae Young, C. (2011, Jun 5). Computational approach to identify enzymes that are potential therapeutic candidates for psoriasis. Enzyme Research. Available at https://pmc.ncbi.nlm.nih.gov/articles/PMC3121017/
31. Etthadi, P., Greaves, M.W., Wallach, D., Aderka, D. and Camp, R.D. (1994 Apr). Elevated tumor necrosis factor-alpha (TNF-alpha) biological activity in psoriatic skin lesions. Clin Exp Immunol. Available at https://pmc.ncbi.nlm.nih.gov/articles/PMC1534536/
32. Morizane, S. and Gallo, R.L. (2012, Dec 20). Antimicrobial peptides in the pathogenesis of psoriasis. J Dermatol. Available at https://pmc.ncbi.nlm.nih.gov/articles/PMC3527011/
33. Leonardi, C.L., Powers, J.L., … Gottlieb, A.B., et al. (2023, Nov 20). Etanercept as monotherapy in patients with psoriasis. The New England Journal of Medicine. Available at https://www.nejm.org/doi/full/10.1056/NEJMoa030409
34. Pahar, B., Madonna, S., Das, A., Albanesi, C. and Girolomoni, G. (2020, Sep 10). Immunomodulatory role of the antimicrobial LL-37 peptide in autoimmune diseases and viral infections. Vaccines. Available at https://pmc.ncbi.nlm.nih.gov/articles/PMC7565865/
35. Heseltine, P.N.R., Goodkin, K., … Atkinson, J.H., et al. (1998, Jan). Randomized double-blind placebo-controlled trial of peptide T for HIV-associated cognitive impairment. JAMA Neurology. Available at https://jamanetwork.com/journals/jamaneurology/fullarticle/773550
36. Delfino, M., Fabbrocini, G., Brunetti, B., Procaccini, E.M. and Santoianni, P. (1992). Peptide T in the treatment of severe psoriasis. ActaDV. Available at https://medicaljournalssweden.se/actadv/article/view/16610
37. Marqus, S., Pirogova, E. and Piva, T.J. (2017, Mar 21). Evaluation of the use of therapeutic peptides for cancer treatment. J Biomed Sci. Available at https://pmc.ncbi.nlm.nih.gov/articles/PMC5359827/