The artificial neural network selects saccharides from natural sources a promise for potential FimH inhibitor to prevent UTI infections

Abstract

The major challenge in the development of affordable medicines from natural sources is the unavailability of logical protocols to explain their mechanism of action in biological targets. FimH (Type 1 fimbrin with D-mannose specific adhesion property), a lectin on E. coli cell surface is a promising target to combat the urinary tract infection (UTI). The present study aimed at predicting the inhibitory capacity of saccharides on FimH. As mannosides are considered FimH inhibitors, the readily accessible saccharides from the PubChem collection were utilized. The artificial neural networks (ANN)-based machine learning algorithm Self-organizing map (SOM) has been successfully employed in predicting active molecules as they could discover relationships through self-organization for the ligand-based virtual screening. Docking was used for the structure-based virtual screening and molecular dynamic simulation for validation. The result revealed that the predicted molecules malonyl hexose and mannosyl glucosyl glycerate exhibit exactly similar binding interactions and better docking scores as that of the reference bioassay active, heptyl mannose. The pharmacokinetic profile matches that of the selected bioflavonoids (quercetin malonyl hexose, kaempferol malonyl hexose) and has better values than the control drug bioflavonoid, monoxerutin. Thus, these two molecules can effectively inhibit type 1 fimbrial adhesin, as antibiotics against E. coli and can be explored as a prophylactic against UTIs. Moreover, this investigation can pave the way to the exploration of the potential benefits of plant-based treatments.

Graphical abstract

Supplementary Information

The online version contains supplementary material available at 10.1007/s40203-024-00212-5.

Introduction

Escherichia coli is the major causative agent of symptomatic urinary tract infections (UTIs), which affect approximately 50% of women. In the differential diagnosis of urogenital tuberculosis (UGTB), the second to third-most prevalent form of extrapulmonary tuberculosis, UTIs rank first in creating hindrance which is a matter of concern. The diagnosis of Mycobacterium tuberculosis (Mtb) is difficult and sometimes challenging due to the presence of various pathogens in urine (Lee 2015). Prevalence of urogenital tuberculosis in patients with a long-term history of UTIs evaluated by drug resistance of uropathogens with ineffective antibacterial therapy. As UTIs frequently mask UGTB, a patient with pyuria and bacteriuria is treated with excessive antibiotics rather than being diagnosed for UGTB (Kulchavenya and Kholtobin 2015). A clinical study had been conducted to estimate the prevalence of UGTB and the microflora spectrum among patients having an extended history of UTIs, leading to the evaluation of the existence of uropathogen drug resistance in individuals receiving inadequate antibacterial treatment for UTIs. In UTI patients who had received less effective antibiotic treatment, the overall UGTB frequency was 25.8%. One of the most prevalent UGTB masks is a UTI and comorbid UTIs and UGTB were detected in 65.1% of cases. When normal antibacterial treatment for UTI is inadequate, especially in areas where the disease is prevalent, UGTB should be investigated (Kulchavenya and Cherednichenko 2018). Mtb and E. coli are major pathogens causing antimicrobial threats to human beings. So, the need of the hour is to find such a molecule that is an effective alternative measure for UTIs and is not a threat to antimicrobial resistance (AMR).

A detailed study of targets found on E. coli has shown that the bacterial adhesin lectin FimH, which is located at the tip of type I fimbriae, has been known to be a therapeutic target since the late 1980s (Sharon 1987). FimH contains N-terminal (mannoside binding lectin) and C-terminal (pilin) domains which facilitate the adhesion of the pathogen to the exterior of the host cell membrane through mannosylated glycoproteins. The word lectin was proposed by Boyd in 1954, derived from the Latin word ‘fegere’ meaning, ‘to pick out or choose’ (Boyd 1963). The adhesion is a major cause and is the primary step in bacterial infection. This process can be prevented by competitively inhibiting mannosylated moieties, as an alternate pathway to antibiotics, unlike inducing bacterial resistance. The anti-adhesive affinities of the inhibitors are still in the nanomolar range against FimH (Hatton et al. 2021). Mousavifar et al. report two C-linked glycomimetic antagonists of FimH in sub-nanomolar affinity (Mousavifar et al. 2023).A detailed understanding of the interactions of the inhibitors with the FimH is a prerequisite for developing new promising candidates and molecular dynamic studies are crucial (Krammer et al. 2018; Dumych et al. 2018).The catch bond mechanism and the allosteric interaction between lectin and pilin domains are vital in switching between high/low-affinity states (Aprikian et al. 2007).Regulation of allosteric interactions can be a major area of research to develop anti-adhesives against FimH (Rodriguez et al. 2013). The binding affinity of FimH to various sugars including mannose and indicated that its binding strength is 15-fold higher than that of fructose (Bouckaert et al. 2005).

The realm of traditional medicine contains a wealth of untapped data that motivates modern medicine to process and scientifically demonstrate its biological activity if any. This is important since it adds to the resources of modern medicine and allows a better understanding of the mechanism. The basic constraint in the development of such a protocol includes the isolation of active molecules which is expected to give exact interaction mechanisms, from the composite molecular structures of traditional medicines. ANN-based Machine learning algorithms like self-organizing maps have been successfully employed in predicting the biological activities of molecules as they could discover relationships through self-organization. E. coli was found to be inhibited by the ethanolic extract of Moringa oleifera leaf (EEML) (Jahan et al. 2022). The active ingredient responsible for its antibacterial activity against E. coli is not yet understood.

Artificial intelligence-enabled machine learning algorithms play a major role in processing a large amount of heterogeneous data of molecular systems. Computer-aided drug discovery and repurposing are gaining momentum with the coordination of artificial intelligence and machine learning (AI/ML), systems biology, genomics, etc., to prioritize novel lead molecules and better drug targets (Schneider and Wrede 1998).An unsupervised machine learning algorithm known as ‘self-organizing map’ (SOM), an artificial neural network-based clustering mechanism is widely used in many pharmaceutical types of research, to signify a relationship between drug discovery and biological activity to that of the structure. This process is broadly used in scaffold hopping, drug repurposing (Schneider et al. 2009; Selzer and Ertl 2006), and identification of the binding sites (Harigua-Souiai et al. 2015). SOM-supervised methods have been reported to be more accurate than standard linear QSAR methods in studies (Xiao et al. 2005; Tetko et al. 1993; KR et al. 2018). Competitive learning methodology with various ligand-based virtual screening scenarios has been explained by Hristozov et al. (Hristozov et al. 2007). The artificial neural network is utilized to create a clustering model using FimH active and inactive molecules. In the present study, this methodology was employed to build the model, since known inhibitors of FimH are mannosides, carbohydrates which are screened to predict potential new molecules of the saccharide family that could be active against FimH. This modeling can transform arbitrary dimensional data into a 2D map by computing higher-dimensional nonlinear data on a similarity/difference basis. This paper will eventually support the mechanism of action for the activity of EEML against E. coli (Jahan et al. 2022).

Materials and methods

Data set

The National Institutes of Health (NIH) offers the free chemical database PubChem, which includes chemical structures for bioactivity screens by their compound identifier numbers defined in PubChem Compounds (Kim et al. 2023). In 2019, the PubChem collection had 16 bioassays with 160 tested compounds for the FimH protein. For modeling analysis, all FimH actives (69) and inactives (6) have been evaluated and obtained from the PubChem database (Supplementary Table 1). Heptyl mannose was employed as a reference ligand since mannose derivatives could interact with FimH, and the protein ‘4LOV’ containing this ligand was taken from the Protein Data bank. The structural and dynamic impact of heptyl mannose-binding on the FimH has been extensively studied (Vanwetswinkel et al. 2014). Representatives from across all classes of available saccharides (~ 1008) from all types (mono, di, oligo, polysaccharides, etc.) were collected as screening molecules from the PubChem library in structure data file (SDF) format (Supplementary Table 2).

Model building

The self-organizing map (SOM), an unsupervised machine learning clustering algorithm in the Canvas module of the Schrodinger suite (Schrodinger, Inc., LLC, New York, USA), was used to build a model using the SDF files of FimH actives and inactives downloaded from PubChem bioassays and to screen the activity of all saccharide data of PubChem collection. The different sets of descriptors like the topological, physicochemical, LigFilter, and QikProp as well as the physicochemical descriptors were generated. A property-based clustering is performed and the program generates the self-organized maps utilizing all the descriptors of the aforementioned categories that were accessible by undergoing artificial neural network-based competitive learning to create the model cluster. The screening molecules, carbohydrates were screened on this model and the clusters were individually analyzed to find the screened and active cells falling into one group.

Molecular docking studies

Protein preparation

The high-resolution three-dimensional X-ray crystal structures of the FimH protein is retrieved from the protein data bank (PDB) (http://www.rcsb.org/) using accession ID 4LOV. The protein was pre-processed in Maestro 9.3's protein preparation wizard (Schrodinger, Inc., LLC, New York, USA). Beyond 5A0 from the hit groups, the metals and water molecules were eliminated and the protein was minimized in the OPLS-3e force field.2.3.2 Ligand preparation.

The conformation structure of the screened carbohydrate molecules was prepared using the LigPrep (Schrodinger, LLC, NY, USA, 2009). The Hyperchem Student evaluation version is used to optimize the compounds that lack 3D structures before the structure is transferred into Maestro to prepare the ligand.

Molecular docking

The docking study was carried out with the Grid-Based Ligand docking method to analyze the interaction of selected carbohydrate compounds with FimH protein. Glide generated the states at a pH of 7.0 ± 2.0 after the protein was pre-processed in Maestro 9.3's protein preparation wizard (Schrodinger, Inc., LLC, New York, USA). Beyond 5Ao from the hit groups, the metals and water molecules were eliminated and the protein was minimized in the OPLS-3e force field. The grid for docking was made using the centroid of the built-in inhibitor, heptyl mannose on the protein 4LOV. The compounds were selected for further investigation based on docking metrics such as docking score and glide energy, as well as the data such as hydrogen bonding interactions.

Toxicity test and ADMET analysis

The toxicity test was conducted using Eli Lilly MedChem rules filter, a set of 275 rules, and an open-source program that runs from the Ubuntu command line (Bruns and Watson 2012). The best-docked molecules from Glide docking results were further subjected to absorption, distribution, metabolism, and excretion (ADME) predictions using the QikProp tool of Maestro. The 50 parameters including molecular weight, hydrogen bonding, partition coefficients, absorption parameters, rule of five, etc., are evaluated. 5 similar molecules selected from the library of 1712 drugs of the database are compared and the similarity is noted for further analysis.

Molecular dynamics simulation

The validation using molecular dynamic simulation of the Desmond module as proposed by the Schrodinger suite was adopted. MD simulations of these molecules were performed for 100 ns. The Protein Preparation Wizard of the Schrodinger suite was initially used to create the protein–ligand complex, with default settings that included hydrogen and bond reassignment, the addition of missing side chain atoms in amino acid residues, loop residue optimization, and water orientation at pH 7.4. A periodic simulation box was made using the System Builder module, the system was solvated using the TIP3P water model and neutralized by adding counter ions, and energy minimization was accomplished using the OPLS with 1000 iterations of the steepest descent technique (all-atom force field; optimized potentials for liquid simulations). A Nose–Hoover thermostat (300 K, relaxation time = 1 ps) and an isotropic Martyna-Tobias-Klein barostat (1.01325 bar, relaxation time = 2 ps) were used to monitor the progress of an unrestrained production phase with NPT ensemble after it had reached equilibrium. Both long-range and short-range Coulomb interactions were assessed using the smooth particle mesh Ewald (PME) method and the RESPA integrator. The frames capturing the system's dynamic movements were exported at a 5 ps period. The stability of the system was examined by plotting histograms for RMSD, RMSF, hydrogen bond analysis, radius of gyration (Rg), and torsional bonds (Dhanalakshmi et al. 2022; Pandya et al. 2022).

Results and discussion

All classes of available saccharides in SDF format from the PubChem repository were collected for the study, as mannosides are the inhibitors of FimH. The ability of mannose to cure UTIs has been thoroughly investigated, and the protein FimH that contains the heptyl mannose ligand is used in the study as a reference ligand (Kim et al. 2023; Dhanalakshmi et al. 2023; Rakhila et al. 2018; Lenger et al. 2020).

The SDF files of FimH actives and inactives were downloaded from PubChem bioassays to generate a model, so as to predict the bioactive saccharides. The saccharides were fed into the model built by SOM, the machine-learning algorithm in Canvas. The 3D pharmacophore descriptors, topological descriptors, and 2D fingerprints generated were used to match the properties and the self-organized clusters of screening saccharides along with which bioassay actives had been created (Table 1).

Table 1.

Molecular clusters of saccharides generated by SOM

Cluster number	Number of molecules	Actives	In-actives	Saccharides	Considered for the study	Reason for selection
1	164	1	__	163	Discarded	Single-active
2	142	__	__	142	Discarded	Only saccharides
3	72	__	__	72	Discarded	Only saccharides
4	70	__	__	70	Discarded	Only saccharides
5	61	1	__	60	Discarded	Single-active
6	30	19	__	11	Selected	Most balanced cluster among all with more actives and fewer saccharides and no in-actives
7	29	6	__	23	Discarded	Saccharides are more only 6 actives

The best cluster (cluster number 6) as indicated in Table 1 is chosen for a second SOM analysis since it has slightly variant molecules according to the colour code and has the highest number of actives and screened saccharides (Fig. 1a). The SOM was generated by property-based clustering and only one was observed which has a maximum number of molecules (4) with an active and three screened saccharides (Fig. 1b).

Fig. 1.

a The cell population of the property-based SOM of Saccharides and bioassay actives/Inactives of FimH. b SOM of the selected cluster number 6 and the molecules in the active cluster

These were further taken for confirmation of their biological activity (active cluster) through ligand docking using Glide from Schrodinger. The docking score of molecules of the active cluster with PubChem CID 12881621, 52,397,428, 91,972,250, and 8,765,311 were −5.724, −5.205, −5.555, and −3.481 respectively, whereas the reference ligand docking score was −5.08. The interactions of the predicted actives and the reference ligand are shown in Fig. 2.

Fig. 2.

Ligand interactions of predicted actives and reference active

The detailed interactions including the hydrogen bonding are tabulated in Table 2 and the predicted actives have the best matching interactions with the reference ligand chosen. The two predicted actives with CID-129881621 and CID-91972250 were Malonyl hexose (MH) and Mannosyl glucosyl glycerate (MGG) respectively, which are naturally found with glycosides of flavanoids and subjected to the ADMET study.

Table 2.

Comparative profile of Lipinski’s rule of five for predicted actives

Molecule	Molecular weight	Rotatable bonds	Hydrogen bond donor	Hydrogen bond acceptor	QPlogPo/w	Rule of five
Reference active	278.345	12	4	10.2	−0.063	0
Predicted active MGG	430.362	16	8	21.4	−3.911	2
Predicted active MH	428.346	16	8	18.6	−3.863	2
Compared BF_QMG	550.429	13	6	15.05	−0.813	3
Compared BF_KMG	654.577	18	9	22.25	−2.325	3
Compared BF_drug monoxerutin	534.429	12	5	14.3	−0.201	3

All three screened saccharide molecules, among the 4 molecules in the active cluster (one bioassay active) were taken for Eli Lilly promiscuity screening. The risk of failure could be reduced if the intervention of off-target effects in the primary stages of drug discovery is properly addressed. The Eli Lilly MedChem rules filter, developed by Eli Lilly, USA from their 18 years of medicinal chemistry experience, profiled sub-structures based on an index of biological promiscuity to reject them from the screening assay (Bruns and Watson 2012). It is widely used for in-silico ADMET and promiscuity profiling to prioritize the best lead molecules without promiscuous activity. Two predicted molecules passed the test, but the molecule which had the lowest dock score was declared as ‘poor’.

The predicted actives were fed into the Qikprop wizard of the Schrodinger suite. Among five drugs of 1712 molecules selected through the process of retrieval of similar molecules, the most similar to MH were Lactulose, Lactitol, Monoxerutin, Hexoprenaline, and Thiamine. Similar drugs listed for MGG were Lactulose, Lactitol, Monoxerutin, Hexoprenaline, and Enprostil. The first two were already present in the saccharides selected for screening and not found along with any actives. The third on the list is the natural flavonoid glycoside Monoxeruthin which was chosen as a reference drug. Tables 3 and 4 list out the values for the new candidate and are comparable with the reference bioassay active molecule.

Table 3.

Comparative profile of ADME parameters for predicted actives

Molecule	SASA	FOSA	FISA	PISA	Volume	QPPCaco	QPlogBB	HOA
Reference active	577.464	409.747	167.717	0	981.353	254.356	−1.794	2
Predicted active MGG	621.972	227.578	394.394	0	1151.449	0.457	−4.004	1
Predicted active MH	608.674	195.198	413.476	0	1123.094	1.188	−4.141	1
Compared BF_QMG	752.291	105.52	465.214	181.557	1431.303	0.097	−4.873	1
Compared BF_KMG	864.162	280.666	417.978	165.518	1714.83	1.077	−4.856	1
Compared BF_Drug_Monoxerutin	743.213	106.485	420.964	215.763	1410.755	0.256	−4.349	1

Table 4.

Ligand interactions of predicted actives along with natural flavonoid counterparts in comparison with heptyl mannose and FimH

Ligand/(Dock score)	Charged negative	Charged positive	Polar	Hydro phobic	Hydrogen bond	Glycine	Salt bridge	Pi-pi stacking
Heptyl alpha-D-Manno pyrannoside (5.08)	ASP 47 ASP 54 ASP140		ASN 46 GLN133 ASN 135 ASN 138	TYR 48 PHE 1 ILE 52 TYR 137 PHE142 ILE13	PHE 1 (1.83, 1.80) ASP 47 (2.02) ASP 54 (1.76, 1.68) GLN133 (2.13) ASN 135 (1.84) ASP140 (2.04)	GLY 14
Mannosyl glucosyl glycerate (MGG) (−5.555)	ASP 47 ASP 54 ASP140		ASN 46 GLN133 ASN 135 ASN 138	TYR 48 ILE 52 PHE 1 TYR 137 ILE 13 PHE142	PHE 1 (1.85, 1.89) ASP 47 (2.08) ASP 54 (1.77, 2.13) GLN133 (2.26) ASN135(2.15, 2.52) ASP140 (1.90)	GLY 14
Malonyl-hexose (MH) (−5.724)	ASP 47 ASP 54 ASP140	ARG 98	ASN 46 GLN133 ASN 135 ASN 138	TYR 48 ILE 13 PHE 1 ILE 52 TYR 137 PHE142	ASP 47 (2.03, 1.77) PHE 1 (1.79) ASP 54 (1.78, 2.05) ASP140 (1.59) GLN133 (2.59) ASN 135 (2.47)
Monoxerutin (−4.726)	ASP 47 ASP 54 ASP143	HIP 45 ARG 98	ASN 46 GLN133 ASN 135	PHE 1 ILE 52 TYR 48 PHE142 ILE 13	ARG98 (2.63, 2.43) ASP47(3-BONDS) (1.86, 2.01, 1.98) ASP 54 (1.81) PHE 1(1.82)			TYR 48 (4.04)
Quercetin 3-O-malonyl glucoside (−5.353)	ASP 47 ASP 54 ASP140	ARG 98	ASN 46 GLN133 ASN 135 ASN 138	TYR 48 ILE 52 PHE 1 ILE 13 PHE142	GLN133(3-BONDS) (1.67, 2.50, 2.50) ASN 135 (2.07) ASP 47 (2.78, 2.40)			PHE 1 (4.27, 3.57)
Kaempferol 3-o-beta-D-(6''-o-malonyl)-glucoside (−5.173)	ASP 47 ASP 54 ASP140	HIP 45 ARG 98	ASN 46 GLN133 ASN 135 ASN 138	TYR 48 PHE 1 ILE 52 TYR 137 PHE142 ILE13	ASP 47 (2.58, 2.34) ASN 135 (1.78) PHE 1 (2.12) GLN133 (1.95)		PHE 1 (4.72) ARG98 (4.77)

As the drug taken for comparison was a flavonoid glycoside and the predicted active Malonyl hexose is found along with bioflavonoids quercetin and kaempferol, their activity along with the two predicted molecules was analyzed first. Later, the values for two predicted actives were recorded and compared with the drug bioflavonoid monoxerutin and other natural counterparts, viz. Quercetin 3-O-malonyl glucoside (BF_QMG) and Kaempferol 3-o-beta-D-(6''-o-malonyl)-glucoside (BF_KMG). Malonyl hexose is found in natural companions of flavonoids of traditional medicinal plants like Moringa oleifera and Cecropia obtusifolia (Makita et al. 2016; Rivera-Mondragón et al. 2019). Among the abundant family of phytonutrients with established therapeutic qualities, flavonoids are one of the most important phytochemicals in diets and are of significant interest because of their wide bioactivity. Many of these activities are ascribed to their glycosidic side-chain moieties (Panche et al. 2016; Slámová et al. 2018; Xiao 2017).

According to the rule of five, for a drug exhibiting good absorption, the number of hydrogen bond donors (HBD) should not be more than 5, the hydrogen bond acceptor (HBA) not to be more than 10, the molecular weight (MW) should not exceed 500 and Log P value not to exceed 5 (Lipinski et al. 1997; Divyashri et al. 2021). Any molecule that obeys this rule is considered drug-like. The maximum violation permitted compared with many properties of similar drugs is 4. The studied molecules have only 2 and the compared drug Monoxerutin has 3, therefore, the actives are qualified to be druglike.

Solvent accessible surface area (SASA) can fall between 300 and 1000 whereas the hydrophobic component of SASA falls in the range of 0.0–750.0, the hydrophilic component of SASA should be between 7.0 to 330.0, and the value of the Pi Component of SASA should be 0.0–450.0, well inside the preferred limits. The solvent-accessible volume, predicted apparent Caco-2 cell permeability—a model for the gut-blood barrier, for non-active transport, and the predicted brain/blood partition coefficient are comparable with the reference drug including human oral absorption (HOA).

In order to examine characteristics of conformational modifications resulting in variations of protein–ligand interactions and protein folding, the molecular dynamics simulation technique has been systematically used. In this study, SOM clustering, molecular docking investigations, toxicity test, and ADMET profiling revealed that MGG and MH molecules can be employed as FimH inhibitors. The molecular dynamics simulation of FimH-MGG and FimH-MH was performed for 100 ns. Figure 3 depicts the RMSD deviations in the protein segments in the simulation, which includes protein equilibration (left Y-pivot). The RMSD examination is required for analyzing the usual change in the movement of structural atoms with respect to the reference beginning frame while evaluating the trajectories of MD reproductions. The simulation's equilibration may be verified using RMSD analysis—its structural modifications at the conclusion of the process are concentrated on a thermally, energetically stable state. For small, globular proteins, changes in the range of 1–4 Å are quite acceptable. However, when the protein's size increases, the RMSD value's range increases. For the complex of FimH-MGG (Fig. 3a), the protein backbone hovers the value of RMSD not exceeding 2.4 Å; for the FimH-MH complex (Fig. 3b), the value stays well under 2.4 Å as well. The term ‘Lig fit Prot’ refers to the RMSD values of the ligand in relation to the protein backbone. For this, values considerably higher than the protein's RMSD are usually acceptable; however, if the values detected are substantially greater than the protein's RMSD, it is probable, that the ligand adopts a modified stable position than the initial posture. For FimH-MGG (Fig. 3a), the Lig fit Prot value stays below 8 Å but after the time frame of 30 to 70 ns, the RMSD slightly increases, and then stabilizes after 80 ns, suggesting the MGG changes poses and then stabilizes. Similarly, complex FimH-MH stabilizes after 50 ns.

Fig. 3.

MD simulation Protein-ligand interaction root-mean-square deviation (RMSD) profile of a FimH-MGG b FimH-MH

The information from the RMSF value about the dynamic behavior of protein in an aqueous simulated system may be used to represent localized minute changes throughout the length of the protein chains. The pinnacles in Fig. 4 represent areas of the protein that change the greatest during the course of the simulation. The N- and C-terminal ends of proteins frequently change more than other regions inside the protein. In general, auxiliary protein regions like alpha helices and beta strands are more rigid and inflexible than the unstructured parts, and as a result, they sway differently from the protein's loop segments. Both complexes, FimH-MGG (Fig. 4a) and FimH-MH (Fig. 4b) depict protein collaboration with the ligands and patterns for the values of RMSF and B-factor compare similarly in this condition. Both protein–ligand complexes are anticipated to have fewer changes iterating a reduced flexing of the protein backbone.

Fig. 4.

MD simulation Protein-ligand interaction root-mean-square fluctuation (RMSF) profile of a FimH-MGG b FimH-MH

Throughout the MD simulation, the interaction between protein and ligand was detected. Its classification is under four categories: water bridges, hydrophobic connections, ionic contacts, and hydrogen bonds which may be investigated using the graphical depiction of a 'Simulation Interactions Diagram.' The stacked bar traces represent normalized cumulative interaction profiles; for instance, an evaluation of 0.8 denotes that the associated interaction is likely to be persistent 80% of the time throughout the simulation. Characteristics greater than 1.0 exist because some amino acids may make more than one form of interaction with the ligand of an equal subtype.

Figure 5a, b show that the interaction pattern seen in the docking data is validated during MD simulation for both MGG and MH (Dumych et al. 2018). The normal associated amino acids in H-bond interaction for FimH-MGG are Phe1, Asp47, Asp54, Arg92, Gln133, Asn135, Gly14, Tyr21, Ser78, Asn96, Val94, Lys101 and amino acids in Hydrophobic interaction are Tyr48, Ile52, Tyr55, Pro91, Val93, Tyr 137. The interacting amino acids in H-bond for FimH-MH are Phe1, Asp47, Asp54, Asn135, Asn134, and amino acids in hydrophobic interaction are Tyr48 and Ile52.

Fig. 5.

Interaction profile of crucial interacting amino acids of the FimH in contact with a MGG b MH; Ligand interaction diagram displaying total time (in %) a particular amino acid of the protein over the course of the simulation

Conclusion

Artificial intelligence-based machine learning approaches should be effectively engineered for the in-silico drug discovery bottlenecks in order to achieve better prediction profiles. It is concluded that the predicted molecules malonyl hexose and mannosyl glucosyl glycerate exhibit exactly similar binding interactions and better docking scores as that of the reference bioassay active, heptyl mannose. The pharmacokinetic profile matches that of the selected bioflavonoids (quercetin malonyl hexose, kaempferol malonyl hexose) and has better values than the control drug bioflavonoid monoxerutin. Thus, these two molecules can effectively inhibit type 1 fimbrial adhesion, as antibiotics against E. coli, the causative agent of UTI and other diseases. This paper is supported by the results of Jahan et al. for the study conducted on the activity of ethanolic extract of Moringa Olifera leaves against E. coli. The present study suggests that malonyl hexose or flavonoids containing malonyl hexose can be the active ingredients responsible for the antibacterial activity, the mechanism being the inhibition of FimH, the lectin present in E. coli.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

ADMET

Absorption, distribution, metabolism, excretion, and toxicity

AI

Artificial intelligence

AMR

Anti-microbial resistance

ANN

Artificial neural networks

EEML

Ethanolic extract of Moringa oleifera leaf

FISA

Hydrophilic component of SASA

FOSA

Hydrophobic component of SASA

HBA

Hydrogen bond acceptor

HBD

Hydrogen bond donor

HOA

Human oral absorption

KMG

Kaempferol 3-o-beta-D-(6''-o-malonyl)-glucoside

MD

Molecular dynamic simulation

MGG

Mannosyl glucosyl glycerate

MH

Malonyl-hexose

ML

Machine learning

Mtb

Mycobacterium tuberculosis

MW

Molecular weight

PISA

Pi component of SASA

QMG

Quercetin 3-O-malonyl glucoside

QP logBB

Predicted brain/blood partition coefficient

QPPCaco

Caco-2 cell permeability coefficient

QSAR

Quantitative structure-activity relationship

RMSD

Root mean square deviations

RMSF

Root mean square fluctuations

rUTIs

Recurrent urinary tract infections

SASA

Solvent accessible surface area

SOM

Self-organizing maps

UGTB

Urogenital tuberculosis

UPEC

Uropathogenic Escherichia coli

UTI

Urinary tract infection

Author contributions

Conception and design were done by MD, guided by NM Andal and SD Simulation studies were done by AG and MP. Toxicity studies conducted by JKR. SD, DS, and KD have reviewed and edited the draft manuscript. All authors have read, and approved the final manuscript.

Data availability

Publicly available.

Declarations

Conflict of interest

The authors declare that they have no known competing financial interests that could have appeared to influence the work reported in this paper.