Introduction

Psoriasis is a persistent immune-mediated inflammatory skin disease significantly impacting patient’s quality of life. The global disease prevalence ranges from 2 to 3%, and in India it ranges from 0.44 to 2.8%1. Psoriasis is often believed to be aggravated by various factors, including stress, excessive alcohol consumption, smoking, infection, and certain drugs such as beta blockers, lithium, angiotensin-converting enzyme inhibitors, and withdrawal of corticosteroids2,3. The hallmark of psoriasis is prolonged inflammation leading to uncontrolled keratinocyte growth and dysfunctional differentiation. The histology of the psoriatic plaque shows acanthosis (epidermal hyperplasia), overlying inflammatory infiltrates consisting of dermal dendritic cells, macrophages, T cells, and neutrophils4.

Treatments include several topical and systemic therapies along with phototherapy. Treatment involves methods to reduce pain and disability caused by arthritis and associated manifestations5. As the exact cause of psoriasis is still unclear, treatment is only available to control symptoms. Moreover, the available biologics for disease treatment, including Etanercept, Adalimumab, and Infliximab, are stated to be associated with a high cost of treatment, a significantly increased risk of serious infections, and have also been reported to show major contradictions in patients with tuberculosis and cardiovascular disorders. Several rare cases of Crohn’s disease and ulcerative colitis were also reported as an adverse effect in clinical trials of these available psoriasis drugs6. Therefore, drug repurposing could be an appealing strategy to find novel treatments for psoriasis given that it may shorten research and development timelines by 3–5 years, minimize costs, and have a high chance of success7. The concept of drug repurposing is a well-established idea, and minoxidil is considered one of the earliest examples of a repurposed medicine. The FDA approved systemic minoxidil for arterial hypertension in 1979 and topical minoxidil for androgenetic alopecia in 1988.

Several epidemiological and “omics” studies found psoriasis to be associated with various other ailments, making it a multi-systemic disorder8,9,10,11,12,13,14. Hence, a systematic literature search was done to cumulate the available evidence of psoriasis associated disorders, including autoimmune disorders, cardiovascular risk factors, etc. (Table 1). Further, based upon the available evidence, an exhaustive database search was done to identify FDA approved drugs along with their targets, which could possibly be used in repurposing as potential anti-psoriasis drugs.

Table 1 Summary of psoriasis associated diseases.
Full size table

Therefore, the present study focused on the repositioning of FDA approved drugs or those in the clinical phase of the associated autoimmune disorder and comorbid conditions using multiple in-silico approaches including network pharmacology-based methods, molecular docking, one-SVM algorithm prediction, and molecular dynamics simulation to propose potential anti-psoriasis drugs.

Methodology

Data retrieval

The known drug target interactions of the diseases- 19 autoimmune diseases, 4 cardiovascular risk factors, and each of infectious, lung, and mood disorders (Table 1) were retrieved using various public databases, i.e., DrugBank, PharmGKB, clinicaltrial.gov database, TTD, CTD, and the Unified Medical Language System NDF-RT. To ensure data integrity, duplicate drug-target interactions were identified and removed using a combination of automated scripts and manual curation. Each drug was assigned a unique PubChem CID, and targets were mapped to UniProt IDs to avoid redundancy. Cross-validation was performed by comparing entries across DrugBank, PharmGKB, and the Unified Medical Language System (NDF-RT), eliminating inconsistencies.

Gene ontology (GO) mapping and drug-target prioritization using NBI score

Molecular function GO mappings from the QuickGO database, a version of EBI was downloaded and subsequently filtered depending upon the information, like the absence of biological data and missing inferences from electronic annotation, etc. The leaf node of the GO annotation was mapped to the curated targets retrieved from various databases. If two targets shared the same GO information, a new DTI edge was curated. The predicted target for each drug was prioritised by sorting each drug with respect to its descending NBI score, calculated as per Cheng et al.39. Subsequently, the top-scoring drug will be considered as prioritised drug target7.

Molecular docking of prioritized drug-target

The drug target interaction of prioritised drugs was analysed using a molecular docking approach.

Ligand preparation

The drug structure in SMILE format of the selected compounds was obtained from PubChem, DrugBank, and PharmGKB databases. 3D and geometry optimizations of ligands were performed using energy minimization utilizing techniques in Schrödinger Maestro v. 11.440. The LigPrep module from Schrodinger, LLC, NY, USA, 2009, was utilized via the Maestro builder panel to produce ligands. This involved generating 3D structures of the ligands by adding hydrogen atoms, eliminating salt, and ionizing at pH (7 ± 2)41. Energy minimization was conducted using OPLS_2005 force field through the standard energy function of molecular mechanics. A root mean square deviation (RMSD) cut-off of 0.01 Ǻ was applied to produce the low-energy ligand isomer42.

Protein preparation

Protein crystal structure was retrieved from the Protein Data Bank (http://www.rcsb.org)43. A standard PDB structure file contains heavy atoms and may contain a co-crystallized ligand, water molecules, metal ions, and co-factors. Some multimeric structures are required to be reduced to a single unit due to the poor resolution of X-ray studies, making the differentiation between nitrogen (NH) and oxygen (O) difficult. The location of these groupings was cross-checked. The key connectivity information, bond ordering, and missing formal charges in the PDB structure were allocated by preparing a downloadable crystal structure for further in-silico investigations. The Schrodinger software’s protein preparation wizard was utilized to eliminate water molecules beyond 5 Å from hetero groups, add hydrogens, manage metals, and apply bond ordering to the protein. The protein was reduced to a minimum level of 0.3 from the original structure using the Impref module of Impact with OPLS 2005 force field. Hydrogens were fine-tuned after thorough sampling as described by44. The precision and robustness of the protein structure were assessed using a Ramachandran plot.

Prediction of active sites and generation of receptor grids

The active site residues for the target were projected based on 3D structure of proteins or literature sources45. The anticipated active sites of the target protein were cross-validated with SiteMap46. A receptor grid was created for the protein structures at the specific binding sites with higher site score47.

Molecular docking

The docking experiment was carried out using standard parameters after preparing the protein and ligand, determining the binding site, and constructing the receptor grid in the Glide module of Schrodinger software48,49. To ensure higher accuracy, the molecular dynamics simulation was integrated with flexible docking experiments. The Glide process in Maestro v11.4 uses grid-based ligand docking (Glide) software to achieve high accuracy flexible docking (FD). The ligands were docked using “Extra Precision Mode” (XP). The docked conformers were assessed based on their Glide (G) Score. The G Score is calculated as follows:

$${\text{GScore}} = {\text{a}}*{\text{vdW }} + {\text{ b}}*{\text{Coul}} + {\text{Lipo}} + {\text{Hbond}} + {\text{Metal}} + {\text{BuryP}} + {\text{RotB }} + {\text{Site}}$$

wherein vdW denotes van der Waals energy, Coul denotes Coulomb energy, Lipo denotes lipophilic contact, H bond indicates hydrogen-bonding, Metal indicates metal-binding, BuryP indicates penalty for buried polar groups, RotB indicates penalty for freezing rotatable bonds, Site denotes polar interactions in the active site and the a = 0.065 and b = 0.130 are coefficients of vdW and Coul.

Binding free energy calculation using prime/MM-GBSA approach

The free energy of binding for lead compounds was calculated using Prime/MM-GBSA. The hapten-protein complex structure created from Glide FD served as the source for the simulation. The complex energies were determined using the OPLS-AA (2005) force field and generalized-Born/surface area (GB/SA) continuum solvent model. The docked positions were refined using Prime’s local optimization function. The binding free energy, Gbind, is computed using Eqs. (1 and 2)50,51.

$$\Delta {\text{G}}_{{{\text{bind}}}} = \Delta {\text{E}} + \Delta {\text{G}}_{{{\text{solv}}}} + \Delta {\text{GSA}}$$
(1)
$$\Delta {\text{E}} = {\text{E}}_{{{\text{complex}}}} {-}{\text{E}}_{{{\text{protein}}}} {-}{\text{E}}_{{{\text{ligand}}}}$$
(2)

whereas, The Ecomplex (Target protein—drug), Eprotein (Target protein), and Eligand (drug), respectively.

Machine learning model

Machine learning approach was performed using Python 3.7. Libraries such as Pandas, Scikit Learn, Sklearn, and plotly were used. One SVM algorithm was used for prediction.

Data preparation

A total of 36 FDA approved psoriasis drugs with its 58 targets along with their XPG and MMGABS score obtained from molecules docking were used for training the model. Another list of FDA approved drugs obtained after drug prioritization with its targets were used as testing dataset.

Model development

The training dataset contains only one class of data. Data was trained using the one-class SVM through Sklearn, Scikit-learn, and Pandas libraries in Python 3.7. The Support Vector Machine (SVM) is a widely-used machine learning technique that relies on a decision plane concept and is mainly utilized for classification purposes. The one-class support vector machine (SVM) technique aims to encapsulate the underlying inliers. The goal is to categorize data into two classes using a decision function: one class is labeled as inliers (positive) and the other as outliers (negative). Parameters as Kernel = linear, gamma is used as the default “scale” and nv value that reflects the training errors and support vectors was used at 0.65.

Inliers and outliers detection

The test dataset was applied to the developed model. To authenticate the model; model accuracy, precision, recall, and F1 score was calculated. The drugs to be used for psoriasis treatment were predicted. The inliers were indicated as ' + 1' and the outliers as '-1' based on the probability prediction score.

Molecular dynamics (MD) simulation

A molecular dynamics simulation was conducted on proteins and the top three protein–ligand complexes using GROMACS-2021.2 for a 100 nano-second duration, employing the CHARMM force field52. Avogadro was utilized to create the mol2 file of the ligand molecule, which was then used in CGenFF53 to generate the topology file for the ligand molecule for the CHARMM force field. The TIP3P water model was employed to solvate the proteins and complexes within a cubic enclosure. 34 sodium ions and 34 chlorine ions were added to neutralize the protein molecule 2BXR, 10 chlorine ions were added to the 6WJC, 6WJC-dimetindiene, and 6WJC-trimipramine complexes to neutralize the protein–ligand complexes. Energy minimization steps in the 2BXR, 6WJC, 2BXR-pioglitazone complex, 6WJC-dimetindiene complex, and 6WJC-trimipramine complex, were 273, 1411, 1119, 1412, and 1681 respectively and the algorithm used for energy minimization was the steepest descent. Non-isothermal-isobaric (NPT) and non-isothermal-isovolumetric (NVT) experiments were conducted at a temperature of 298 K and a pressure of 1 bar. The production molecular dynamics (MD) simulations were conducted at a 100 nano-second timescale for all systems. The MD trajectories of the protein, as well as complexes, were further analyzed by root mean square deviation (RMSD), root mean square fluctuation (RMSF), radius of gyration (Rg), solvent accessible surface area (SASA), intermolecular hydrogen bonds, intramolecular hydrogen bonds, and minimum distance using various modules of GROMACS. The minimum distance values of the complexes were subsequently examined to study the molecular interactions between protein and ligand molecules. Regression analysis was conducted on the minimal distance between protein and ligand molecules to analyze the molecular interaction pattern throughout a 100 ns timeframe. The MASS package in RStudio was utilized for regression analysis54.

Binding free energies calculation of MD trajectories

MD simulations were utilized to conduct MM/PBSA binding energy calculations55 in order to characterize the protein–ligand interactions. The free energy of a state (ligand, protein, or complex) is calculated in MM/PBSA by adding the following sums:

$${\text{G }} = {\text{ Ebad }} + {\text{ Eel }} + {\text{ Evdw }} + {\text{ Gpol }} + {\text{ Gnp }} - {\text{ TS}}$$

where the first three components are usual MM energy terms, Ebad is derived from binding (bond, angle, and dihedral), Eel indicates electrostatic interactions, and Evdw implies van der Waals interactions, respectively. The polar and nonpolar contributions to the solvation free energy are denoted by Gpol and Gnp, respectively. Gpol is commonly calculated using the Poisson–Boltzmann equation or the generalized Born (GB) model, whereas the nonpolar component is approximated using a linear relationship to the solvent accessible surface area. The last term is the absolute temperature, T, multiplied by the entropy, S, which is deliberate using a normal-mode exploration of the vibrational frequencies.

Results

Data retrieval

A systematic literature search identified; 19 autoimmune disorders, 5 cardiovascular risk factors, 2 inflammatory conditions and a mood disorder closely aligned to psoriasis (Table 1). The FDA approved drugs were explored using DrugBank, PubChem and PharmGKB databases. Based upon the database search, a total 517 drugs and 2478 targets in respect to these drugs were retrieved. Further, 36 drugs being widely used to treat psoriasis were retrieved and used as standard (Supplementary Table I).

Promiscuity in drug-target interactions and gene ontology based enriched drug-target interaction network

A drug is promiscuous when it acts on multiple targets and exhibits distinct pharmacological effects. The study instated drug promiscuity of nearly 92% (476 drugs) with < 10 targets, 5.6% (29 drugs) with < 20 targets, 1.1% (6 drugs) with < 30 targets, 0.58% (3 drugs) with < 40 targets, 0.38% (2 drugs) with < 50 and 1 drug i.e., Fostamatinib (DB12010) with 300 targets (Supplementary Table II). On combining the Drug-Bank DTIs and GO mapping DTIs, 7063 edges with 2193 unique targets with respect to 517 drugs were further used to determine the NBI score.

Identification of new drug-target interactions using network based inference

For the entire network, comprising of 517 drugs and 2193 targets NBI association scores were computed. A prioritization list was created where a target was only prioritized for a drug if its association score was more than 20% of the maximum score in the sorted list with unconnected targets. This led to identification of 358 potential DTIs (211 unique drugs & 119 unique targets), prioritized on their NBI association scores (Supplementary Table III).

Molecular docking of prioritized drug-target

Protein and ligand preparation

A total of 211 drugs & 358 targets were selected for molecular studies. The structures of the compounds were retrieved in SMILE format using PubChem, DrugBank and PharmGKB database. After filtration i.e., removal of duplicates, unavailability of x-ray crystallographic structure, mutated structure and low-resolution structures, 120 drugs with 159 targets were selected for further docking studies (Supplementary Table IV). Further, the known drugs being used in psoriasis treatment were utilized as standard drugs and was docked with the selected targets (Supplementary Table IV). The top selected targeted proteins 2BXR and 6WJC was evaluated for its stability and accuracy using Ramachandran plot. The protein model is believed to be a good quality model if have over 90% residues in the most favoured regions. Hence, the retrieved structure from RCSB-PDB was prepared in protein-preparation wizard of Schrodinger suite to make it stable and accurate. The modelled structure 2BXR have 90.7% residues in most favoured region [A, B, L], 8.3% additional allowed regions [a, b, l, p], 0.3% generously allowed regions [~ a, ~ b, ~ l, ~ p] and 0.8% disallowed regions [XX] in comparison to 88.2% residues [A, B, L], 10.1% [a, b, l, p], 0.9% [~ a, ~ b, ~ l, ~ p] and 0.8% [XX] in the same region of original structure. Likewise, 6WJC have 94% residues in most favoured region [A, B, L], 6% in additional allowed region [a, b, l, p] in contrast to 92% residues in [A, B, L] and 8% in [a, b, l, p] (Fig. 1).

Fig. 1
figure 1

(a) represents crystal structure of 2BXR & (b*) represents Ramachandran plot for modelled 2BXR protein (c) represents crystal structure of 6WJC (d*) represents Ramachandran plot for modelled 6WJC protein. * Most favoured region [A, B, L], Additional allowed regions [a, b, l, p], Generously allowed regions [~ a, ~ b, ~ l, ~ p] and Disallowed regions [XX].

Full size image

Prediction of active sites and generation of receptor grids

Active-site residues in a protein, including hydrophobic amino acids, may contribute to catalysis through critical interactions that position the reacting molecule, organize hydrogen-bonding residues, and define the electrostatic environment of the active site56. SiteMap, a quick and effective application to determine probable protein binding pockets was utilized for identification of the binding sites of all drug-targets. SiteMap utilizes binding site characteristics via creative search and analyses each site by computing parameters such as size, volume, amino acid exposure, and enclosure contact, hydrophobicity, hydrophilicity, and donor/acceptor ratio (Supplementary Table IV). As a result, the binding site for all targets were calculated. The calculated binding score for top targeted protein i.e., 2BXR and 6WJC were (site1 1.076, site2 0.872, site3 0.829, site4 0.73 and site5 0.532) and (site1 1.135, site2 1.03, site3 0.99, site4 0.91 and site5 0.761) respectively. Therefore, based upon site score site1 for the both the protein targets was utilized for further docking studies. Ile (19, 23, 180, 207, 273, 281, 325, 335); Gly (20, 21, 22, 25, 49, 50, 66, 67, 71, 110, 214, 434, 443); Ser (24, 209, 403); Leu (42, 97, 176, 259, 277, 337); Glu (43, 216, 323, 446); His (242); Ala (44, 68, 272, 448); Arg (45, 47, 51, 109, 206); Asx (48); Thr (52, 73, 169, 204, 205, 245, 336, 435); Tyr (69, 106, 121, 124, 402, 407, 444); Val (65, 70, 91, 93, 98, 101, 210, 244, 303); Pro (72, 107, 118, 243, 274); Gln (74, 99, 215, 401); Phe (108, 173, 177, 208, 352); Trp (116,128, 397, 441); Asn (117, 181); Lys (280, 305); Met (324, 350, 445); Cys (406) are putative amino acids in the active site region of 2BXR within 4 A° area57. Likewise; for 6WJC protein Val (25, 113, 385); Ile (72, 180); Tyr (82, 85, 106, 179, 189, 192, 381, 404, 408); Leu (86, 102, 183); Gly (89); Trp (101, 157, 378, 400); Asp (105); Ser (109, 184, 388); Gln (110); Cys (178, 407); Pro (186); Ala (193, 196); Phe (197), Asn (382); Thr (389); Glu (397, 401) were identified as putative amino acids in the active site region within 4 Å area58. Therefore, grids were generated by selecting these residues as centroid amino acid residues.

Extra precision docking and binding free energy

A total of 120 retrieved drugs were docked with 159 target protein to ascertain their binding mode and binding affinity. The docking scores and binding free energies of the lowest energy pose of the drug with its target in active sites on Chain A of the proteins x-ray crystal structures were computed using Maestro in Schrodinger (v11.4) after the undesired ligands and amino acids were removed. The docking results revealed highest binding affinity of − 13.507 kcal/mol for Trimipramine, a tricyclic antidepressant, but with more anti-histaminic and sedative properties as compared to the Anthralin, an anti-psoriatic agent with binding affinity of − 9.465 kcal/mol with its target 6WJC. No hydrogen bond was present among both drug-target interaction i.e., Trimipramine-6WJC & Anthralin-6WJC. Tyr106, Trp157, Phe197, Ala196, Ala193, Tyr381, Trp378, Tyr404, Cys407, Tyr 408 recorded hydrophobic interaction with active site residues in both trimipramine and anthralin. Additionally, the hydrophobic interaction in trimipramine with active site residues was also observed at Ile74, Val113, Val385 and Leu183. Likewise, pi-pi stacking (Tyr381), pi-cation (Trp381, Trp378, Tyr404) and salt-bridge (Asp105) interaction was observed in trimipramine with 6WJC. Hence, these amino acids are believed to have a critical and direct role in protein- drug binding.

Subsequently, the binding affinities was followed by dimetindene (− 12.017 kcal/mol) an antihistamine/anticholinergic drug as compared to the Anthralin, an anti-psoriatic agent (− 9.465 kcal/mol) with its target 6WJC and Pioglitazone (− 12.745 kcal/mol) an antihyperglycemic in comparison to tofacitinib (− 9.689 kcal/mol), a Janus kinases inhibitor affecting haematopoiesis and immune cell function with its target 2BXR (Table 2).

Table 2 Docking based molecular interaction of Top 3 drugs-protein with respect to its standard.
Full size table

The compounds’ MMGBSA binding energies (Kcal/mol) were computed and the binding energies of top 3 drug-targets are presented in Table 2. The prime MM-GBSA method confirmed the stability of drug and protein complex. The triumph molecules from XP docking studies displayed higher binding energies compared to their respective standards with Pioglitazone-Amine oxidase [flavin- containing] A (2BXR) showing the highest ΔGbind − 86.364 kcal/mol in comparison to its standard complex Tofacitinib- Amine oxidase [flavin- containing] A (2BXR) [ΔGbind − 39.433 kcal/mol]. The interaction between top 3 protein-drugs & its standard is depicted in Fig. 2.

Fig. 2
figure 2

Molecular interaction between drug & receptor (a) 2BXR-Tofacitnib [Standard Drug] (b) 2BXR-Pioglitazone (c) 6WJC-Anthralin [Standard Drug] (d) 6WJC- Dimetindene (e) 6WJC- Trimipramine.

Full size image

One-SVM machine learning model

The developed model was adjudged for its authenticity with accuracy of 0.0065, precision value 0.5000, recall value 0.3270 and F1 Score of 0.3954. The One-SVM model predicted the top inlier drugs with anti-psoriasis activity. The inliers validated the docking outcome with top 5 drugs predictions being pioglitazone, dimetindene, trimipramine, amitriptyline and verapamil (Fig. 3). The model also predicted 83 drugs with its 104 targets as outliers (Supplementary Table V).

Fig. 3
figure 3

One SVM-algorithm drug -target prediction Outcome (a) Top 10 predicted inlier potential drug-target (b) Top 10 predicted outlier drug -target.

Full size image

Molecular dynamics simulation and regression analysis

Following the docking calculations, MD simulations on the potential binding site were carried out by using the GROMACS-2021.2. MD analysis measured 2BXR-Pioglitazone, 6WJC-Trimipramine and 6WJC-Dimetindene to determine the accuracy of the proper dynamical binding pattern of these systems throughout the 100 ns MDS trajectories. The physical stability of both systems was analyzed by considering the overall potential energy, RMSF, RMSD, hydrophobic interactions and water mediated patterns.

RMSD plot has shown the changes observed in the protein–ligand complexes as compared to the native protein structure and to ascertain that psoriasis drug candidates maintain their interaction with the target protein throughout the simulation. RMSD of 2BXR-Pioglitazone system followed almost similar 501, G5pattern upto 35 ns, however the native protein (2BXR) has shown higher RMSD value after 40 ns till the end of the trajectory at 100 ns time scale (Fig. 4a). Similarly, RMSD of 6WJC-Trimipramine system followed almost uniform pattern upto 40 ns; though, the unbound protein (6WJC) has shown high RMSD value after 43 ns till the end of trajectory at 100 ns time-scale (Fig. 5a). However, the complex 6WJC-Dimetindene followed almost similar pattern upto 38 ns and the complex have shown higher RMSD value after 53 ns till the end of the trajectory at 100 ns timescale (Fig. 6a).

Fig. 4
figure 4

MD simulation result of 2BXR and 2BXR-Pioglitazone Complex (a) RMSD (b) RMSF (c) Radius of Gyration (d) SASA (e) Eigen value. The red color indicates complex and blue refers to standard.

Full size image
Fig. 5
figure 5

MD simulation result of 6WJC and 6WJC- Trimipramine Complex (a) RMSD (b) RMSF (c) Radius of Gyration (d) SASA (e) Eigen value. The red color indicates complex and blue refers to standard.

Full size image
Fig. 6
figure 6

MD simulation result of 6WJC and 6WJC- Dimetindene Complex (a) RMSD (b) RMSF (c) Radius of Gyration (d) SASA (e) Eigen value. The red color indicates complex and blue refers to standard.

Full size image

RMSF computes the fluctuations observed in each amino acid residue of the protein and protein–ligand complexes and to establish whether psoriasis drug candidates interact with stable protein regions to avoid off-target effects. RMSF value of 2BXR was higher as compared to complex at residues Pro499, Asn494, ArgR493, Phe490, Asx498, Leu491, Thr489, Glu492, Leu495, Thr487. 2BXR-Pioglitazone complex has shown higher fluctuations at residues Thr505, Isl504, Lys503, Leu502, Leu501, Gly500, His488, Ser497, Pro496 and Glu485 (Fig. 4b). Similarly; higher RMSF was recorded at Met1, Glu2, Gly3, Asp4, Met156, Ser5, Asn157, Tyr6, His7 and Val155 in native state of 6WJC. The 6WJC-Trimipramine complex depicted higher variations at Ser445, Leu444, Tyr443, Gln442, Trp251, Trp400, Arg252, Ser153, 40Glu1 and Leu399 (Fig. 5b).

Likewise, in case of 6WJC-Dimetindene complex; higher RMSF in 6WJC was recorded at His7, Thr10, Gln442, Tyr404, Ala441, Trp405, Leu406, Phe72, Ala71 and Leu432 and the 6WJC-Dimetindene complex represented fluctuations at Met1, Glu2, Glu3, Asp4, Ser445, Met156, Ser5, Asn157, Leu444 and Met241 (Fig. 6b).

The radius of gyration value depicts the compactness of the protein molecule in its native as well as complex form and to ensures the protein does not undergo unwanted structural collapse or expansion upon binding to the drug. Both 2BXR and its complex 2BXR-Pioglitazone showed almost similar till 70 ns, little deviation was observed in 2BXR-Pioglitazone complex between 70 and 82 ns and again followed a similar pattern till the end of trajectories (Fig. 4c). Likewise, 6WJC and its complexes 6WJC-Trimipramine (Fig. 5c) and 6WJC-Dimetindene showed similar pattern of trajectories (Fig. 6c).

SASA value determines the folding, unfolding, and stability of the structure of the protein in the complexes as compared to the protein in the native state. The SASA value of the complexes: 2BXR-Pioglitazone, 6WJC-Trimipramine and 6WJC-Dimetindene was higher as compared to its native structure till 100 ns (Figs. 4d, 5d, 6d). Higher eigen values of 2BXR, 2BXR-Pioglitazone complex, 6WJC, 6WJC-Dimetindine and 6WJC- Trimipramine were 15.97, 15.76, 13.14, 17.55, 14.94 respectively (Figs. 4e, 5e, 6e).

Regression analysis

In order to comprehend the pattern of molecular interactions, the minimum distances between the protein and ligand molecules were assessed. In the 2BXR-pioglitazone complex, regression analysis of the minimum distance between the protein and ligand molecule has demonstrated that the ligand molecules have exhibited a robust interaction in the timeframes ranging from 55 to 100 ns (Fig. 7a). Strong molecular interactions were observed between various timeframes at 18–25 ns, 38–55 ns, and 65–100 ns in the 6wjc-dimetindine complex (Fig. 7b). The 6wjc-trimpramne complex has demonstrated a robust interaction between protein and ligand molecules from 01 to 100 ns (Fig. 7c).

Fig. 7
figure 7

Regression analysis outcome for (a) 2BXR-Pioglitazone complex (b) 6WJC-Dimetindine complex (c) 6WJC-Trimpramne complex.

Full size image

Binding energy of drug (Ligand)-protein complex

The free binding energy of top 3 ligand (drug) molecule with protein molecule were calculated using Poisson-Boltzmann & Surface area method. EPB energies of 2BXR-Pioglitazone; 6WJC-Dimetendine and 6WJC-Trimipramine complex for last 500 timeframes were − 14.63 kcal/mol followed by − 11.87 kcal/mol and − 10.07 kcal/mol.

Discussion

Psoriasis is an immune disorder majorly characterized by cutaneous manifestation. The disease although is devoid of any permanent remission but recent years have witnessed the disease management focussing more on reducing systemic inflammation instead of merely lesion skin improvement. To ascertain the similar approach, the study aimed to predict the drug options that could be repurposed as anti-psoriasis agent.

The study initially utilized a data-driven strategy to select targets and identify the existing drugs that could potentially act on psoriasis. The study further used the idea of resource allocation among linked nodes (drug-target pairings) to determine association scores that indicate the level of association between medications and targets that were previously disconnected. This concept was applied to discover a medicine that is an unidentified target for psoriasis and can be repurposed for psoriasis based on NBI score. Association scores were calculated for the network consisting of 517 medicines and 2193 targets. We created a prioritizing list by applying a metric that only prioritized a target for a medicine if its association score exceeded 20% of the maximum score in the sorted list of unconnected targets. 358 possible drug-target interactions (DTIs) were identified, including 211 unique medicines and 119 unique targets. These interactions were ranked based on their NBI association scores. After filtration, 120 retrieved drugs were docked with 159 targets to determine its binding mode and binding affinity. To scrutinize further one-class support vector machine (SVM) algorithm was employed on the dataset to validate the docking outcome. The outcome predicted top 3 drug-targets which included Pioglitazone-2BXR, Trimipramine-6WJC and Dimetindene-6WJC.

Pioglitazone is thiazolidinedione group of oral antidiabetics which work by binding to proliferator activated receptors (PPARs) that regulated carbohydrate and lipid metabolism. PPARs are present on epidermal keratinocytes, and when these receptors are activated, they have been demonstrated to reduce the growth of psoriatic human keratinocytes in a laboratory setting. Due to its pro-differentiating and anti-proliferative effects on keratinocytes, pioglitazone may offer significant advantages in treating skin and metabolic issues associated with psoriasis59,60,61. Singh and Bansali62 conducted a randomized placebo control research showing that insulin sensitizers improved markers of metabolic syndrome and disease severity in psoriasis patients. Likewise; Mittal et al.63 in another RCT concluded addition of pioglitazone to acitretin therapy enhances anti-psoriatic efficacy with minimal side-effect. However; the studies regarding role of pioglitazone in psoriasis are in very nascent stages and further studies are warranted to construct its therapeutic role in psoriasis.

Trimipramine is a tricyclic antidepressant prescribed for serious (endogenous) and reactive (exogenous) depression. The drug is believed to act by enhancing serotonergic and dopaminergic neurotransmission64. However, certain pharmacological properties of anti-depressant are not related to its anti-depressant activity. One such property of TCAs is that it enables to be potent dermatological agent by virtue of H1 and H2 anti-histaminic and anticholinergic properties. Dermal blood vessel possesses both H1 and H2 histamine receptors and combined H1 and H2 antihistamine therapy is believed to be more effective therapy than H1 anti-histamine alone65. Several researchers used trimipramine in other dermatological conditions like chronic urticaria and atopic dermatitis but the studies included very small sample size65,66,67,68,69. Likewise using machine learning and network framework approach Gilvary et al.70 predicted trimipramine as a new indication on Parkinson’s disease with same target 6WJC.

The other predicted drug-target as an indication for psoriasis in the present study was Dimetindene-6WJC. Dimetindene is a first generation selective H1 antagonist which is used as topical antipruritic and orally to treat allergies71. In an RCT, the usage of dimetindene significantly reduced the itch in atopic dermatitis patients72. Similarly, Lever et al.73 in an another RCT established that topical dimetindene may aid in condition mediated through histamine release.

There are several studies that has predicted different drugs which can be repurposed for psoriasis or psoriasis like inflammation. Thatikonda et al.74 corroborated drug repurposing of Niclosamide has the potential to counteract the pathogenicity of psoriasis due to its potent in vitro and preclinical potential in alleviating psoriasis‐like skin inflammation.

In another study, Zhan and Chen75 involving the construction of a candidate genome-wide genetic and epigenetic network (GWGEN) through big data mining, followed by the identification of real GWGENs of psoriatic and non-psoriatic using system identification and system order detection methods, selected Naringin, Butein, and Betulinic acid and combined them with the multiple-molecule drug to target multiple biomarkers of the pathogenesis of psoriasis. Likewise, Nanda et al.76 in their study utilised GWAS and chemogenomic approach to identify drug pandel against the target POLI which showed desirable molecular and toxicity properties to be considered as a therapy for psoriasis. In another study, Ibezim et al.77 used the molecular docking approach which aided in identification of 12 FDA approved drug against 15 studied anti-psoriatic targets.

Conclusion

To the best of author’s knowledge, the present study is one of its kind utilizing multi-dimensional computational approach (network pharmacology-based methods, molecular docking, one-SVM algorithm prediction, and molecular dynamics simulation) to identify the drugs with anti-psoriasis indications. To conclude, the study identified Pioglitazone, Trimipramine and Dimetindene as top three contender amongst many other drugs as a new indication against psoriasis. The major limitation of the present is that these robust computational findings could not be validated in-vivo or. These top drugs could be utilized on imiquimod induced murine psoriasis model for preclinical evaluation and to determine its sensitivity and therapeutic potential which could be a potential new start in psoriasis drug development. Further, this model could be replicated for drug repositioning of FDA approved drugs to propose potential therapeutics for other diseases.