Polyethyleneimine-Folic acid modified manganese coordination polymer delivery platform for artemisinin in the treatment of non-small cell lung cancer

2.1. Chemicals and measurements

Separation was carried out by octadecylsilane (ODS) gel chromatography (ODS-A, 12 nm, S-50 μm, YMC Co., Ltd., Kyoto, Japan) and silica gel chromatography (800-100/200-300 mesh, Qingdao Marine Chemical Co., Ltd., Qingdao, China), and other types of column chromatography. Hypersil GOLD C18 column (5 μm, 21.2 × 250 mm, Thermo) was utilized to prepared preparative HPLC (2535-2489-2414, Waters, Milford MA, USA). The measurement of NMR spectra was conducted through a Bruker DPX 400 MHz NMR spectrometer (Bruker SpectroSpin, Karlsruhe, Germany) utilizing TMS as an internal standard, including both 2D-NMR and 1D-NMR spectra. On an ORBITRAP FUSION mass spectrometer (Thermo, Waltham, MA, USA), high-resolution electrospray ionization mass spectrometry (HR-ESI-MS) analysis was implemented. All the solvents (Tianjin Damao Chemical Reagent Co., Ltd., Tianjin, China) were of analytical grade.

2.2. Isolation and analysis of the compounds

The Artemisia annua plants were sourced from Tenglong Ecological Agriculture Co., Ltd. in Youyang County, Chongqing. This company specializes in cultivating medicinal plants under controlled conditions to ensure consistent quality and traceability. 70% ethanol was utilized to extract 1 g of aerial components of Artemisia annua. Compound 1 was isolated by further separating the 70% ethanol extract (272 g) utilizing a variety of organic solvent extraction techniques, preparative high-performance liquid chromatography (HPLC), and column chromatography (CC). HR-ESI-MS analysis and NMR spectroscopy were exploited to determine the structures of these compounds.

2.3. The synthesis and characterization of CP1

A mixture comprising 0.2 mmol of Mn(NO₃)₂-6H₂O, 0.1 mmol of H₃L ligand, and 10 mL of deionized water was placed in a 15 mL stainless steel vessel lined with Teflon and held at 120°C for 60 h. A small number of translucent crystals, known as CP1, developed after the mixture naturally cooled to ambient temperature. On the basis of the H₃L ligand, the yield was 46.2%. Analytical values for C₈₄H₅₆Mn₅N₄O₂₈: calculated Mn=14.90, H=3.06, C=54.71; found Mn=14.86, H=3.11, C=54.68. Fourier-transform infrared spectroscopy (cm^-1, KBr): 531(w), 727(s), 735(m), 861(m), 1112(w), 1123(m), 1297(w), 1334(vs), 1421(m), 1528(m), 1541(s), 1665(s), 1782(m), 1986(s), 2933(w), 3208(w), 3437(m). Table 1 displays the detailed crystallographic data for CP1.

2.4. The synthesis of CP1@1 and PEI-FA@CP1@1

To encapsulate compound 1 into the Mn-based coordination polymer CP1, CP1 (0.05 g) and compound 1 (0.02 g) were each dissolved in 5 mL of anhydrous ethanol. The solution of compound 1 was gradually added dropwise into the CP1 dispersion under ultrasonic agitation for 5 h to promote efficient adsorption. The resulting composite (CP1@1) was collected by centrifugation at 4000 rpm for min, washed thoroughly with deionized water to remove unbound molecules, and subsequently freeze-dried. To quantify the drug loading performance, encapsulation efficiency (EE) was calculated using UV-vis spectrophotometry. The concentration of unencapsulated compound 1 in the supernatant was determined by measuring absorbance at its characteristic maximum (λ_max = 325 nm), based on a pre-established calibration curve. The EE was calculated using the following Eq. (1):

W_total​ is the initial amount of compound 1 (20 mg), and W_free​ is the amount of compound 1 remaining in the supernatant. The encapsulation efficiency of compound 1 in CP1 was determined to be 78.6%, indicating effective incorporation of the bioactive molecule within the porous coordination framework.

Subsequently, PEI (0.2 g) and FA (0.15 g) were dissolved in 3 mL of DMSO. To activate the carboxyl group of FA, N,N′-dicyclohexylcarbodiimide (DCC, 0.1 g) and 4-dimethylaminopyridine (DMAP, 0.05 g) were added, and the mixture was stirred at 80°C for h. After the reaction, the activated solution was conjugated with the previously obtained CP1@1 through electrostatic interaction and amide bond formation. To eliminate impurities, the product was subjected to dialysis using a membrane (MWCO 3000 Da) for 48 h, followed by centrifugation and freeze-drying.

2.5. Drug treatment experiments

H1975 (ATCC, USA), the NSCLC cell line, was inoculated into 96-well plates (5×10³ cells per well) and cultivated for 24 h. After 24 h of incubation, the cell culture medium was discarded and substituted with PEI-FA@CP1@1 for another 24 or 48 h. CCK-8 reagent (Beyotime, China) was subsequently added, and the absorbency at 450 nm was detected in accordance with the manufacturer’s instructions. Concurrently, treatment of H1975 cells was performed utilizing PEI-FA@CP1@1 for 48 h. Afterward, the extraction of total cellular RNA was conducted, which was reverse transcribed to produce cDNA, and later, the mRNA level of the pro-apoptotic gene Bax was determined through real-time PCR.

2.6. Simulation details

In the current study, we selected the crystal structure of PI3Kα, with PDB ID 4JPS, as the basis for molecular Docking. This choice is based on the critical role of the PI3K signaling pathway in various cancers, making structural information essential for drug discovery. From the Protein Data Bank (PDB), the crystal structure was downloaded, and we carefully examined its resolution and quality to ensure data accuracy and reliability. To conduct the molecular docking simulation, we used AutoDockTools (version 1.5.7) for preprocessing the protein and ligand. This process involved removing water molecules, adding hydrogen atoms, and optimizing the protein conformation to provide an optimal starting structure for subsequent docking. Molecular docking itself was implemented via AutoDock (version 4.2.6), employing the Lamarckian genetic algorithm to probe 50 potential binding poses, thereby identifying the most promising binding sites and modes. In terms of generative machine learning, we employed MolDQN as the generative engine to modify potential drug molecules based on known benchmark compounds. MolDQN is a model based on deep reinforcement learning, capable of efficiently generating new compounds that are structurally similar to target drugs by learning from molecular structural features and their biological activity. Our model was trained on a rich dataset comprising known drugs and their activity data, which included not only molecular structure information but also pharmacological properties and clinical trial results, thereby providing a solid foundation for the model. In the MolDQN model, we used quantitative estimates of drug-likeness score (QED), synthesizability score (SA), along with binding score as parameter weights to ensure that the generated compounds demonstrate optimal performance in terms of activity, synthetic feasibility, and drug properties. Effective binding energy estimation was conducted using Qvina (version 2.1), considering only one binding mode during this process to enhance computational efficiency and accuracy. The SA and QED scores were evaluated through the Gym-molecules library and RDKit. The SA score reflects the synthetic complexity of the compounds, while the QED score assesses drug similarity; together, they enable effective screening of candidates with greater clinical potential. These measures ensure that the potential drugs we generate are not only theoretically viable but also have a high success rate in practical applications.

3.1. Structure elucidation of new compound 1

We used NMR characterization to confirm the isolated chemical 1’s molecular structure. The reference signal for compound 1’s ¹H NMR spectrum (Figure 1) was the residual solvent peak at 7.26 ppm, which was recorded at 400 MHz in CDCl₃ at 25°C. The singlet at δ = 0.89 ppm is associated with the methyl (–CH₃) group. The polycyclic structure at δ = 1.11, 1.24, 1.74, 2.23, 2.49, 2.52, and 2.81 ppm is ascribed to methylene (–CH₂–) groups. Two doublets were found to be olefinic groups at δ = 5.34 and 5.70 ppm. The expected ratios for each type of proton were obtained by integrating these signals. The observed chemical shifts and coupling patterns closely match the compound 1’s simulated structure.

Close

Figure 1.

¹H NMR spectrum of compound 1 (recorded at 400 MHz, in CDCl₃ solvent, at 25°C).

Export to PPT

Subsequently, under the same circumstances, we characterized compound 1 utilizing ¹³C NMR (Figure 2a). We utilized DEPT-135 NMR spectroscopy to reveal the nuclear magnetic resonance signals of compound 1 in CDCl₃ at 400 MHz and 25°C measurement conditions to more accurately identify the carbon atoms in compound 1 (Figure 2b). The spectrum exhibited several characteristic peaks: positive peaks at δ = 18.93, 21.47, 34.82, 40.48, and 56.27 ppm representing CH₃ and CH carbons, and negative peaks at δ = 22.33, 27.76, 34.45, 38.16, and 40.05 ppm corresponding to methylene CH₂ carbons. The clear distinction between the positive and negative peaks allowed us to differentiate CH₃, CH, and CH₂ carbon atoms, which are highly compatible with the modeled structure of compound 1. The general characteristics of the spectrum, such as its excellent resolution and crisp peak shapes, further supported the molecular structure of the sample. These results provide a strong basis for further investigation and offer crucial support for compound 1’s structural characterization.

Close

Figure 2.

(a) Compound 1’s ¹³C NMR spectrum; (b) Compound 1’s DEPT-135 spectrum, positive peaks indicate CH₃ and CH carbons, and negative peaks indicate CH₂ carbons. (Recorded at 400 MHz, in CDCl₃ solvent, at 25°C).

Export to PPT

To determine the correlation between protons directly connected through J-coupling, we used ¹H–¹H COSY 2D NMR spectroscopy, which is a key tool for structural analysis and molecular confirmation. COSY spectra indicate proton autocorrelation through diagonal peaks and coupling between neighboring protons through cross peaks. Figure 3(a) shows that the NMR hydrogen signals on both sides of the diagonal exhibit good symmetry, indicating strong coupling between adjacent protons. In Figure 3(b), the blue and red cross peaks reveal the weak and strong coupling connections between the protons in the sample, helping to identify adjacent and long-range couplings, thus providing comprehensive information about the compound 1’s molecular structure.

Close

Figure 3.

Compound 1’s 2D NMR spectra. (a) ¹H-¹H COSY spectrum. Red horizontal peaks indicate direct coupling between protons; (b) ¹H-¹H COSY spectrum. Blue and red horizontal peaks reveal coupling between neighboring protons; (c) ¹H-¹³C HSQC spectrum. The red elliptical dots represent the direct coupling between the carbon and hydrogen atoms; (d) ¹H-¹³C HMBC spectrum. The red spots stand for multiple bond couplings between carbon atoms and protons, exhibiting long-range coupling information. (Recorded at 400 MHz, in CDCl₃ solvent, at 25°C).

Export to PPT

For further analysis of the direct coupling between hydrogen (1H) and carbon (13C) in compound 1, we conducted ¹H-¹³C HSQC (Heteronuclear Single Quantum Coherence) spectroscopy. The measurement conditions were a 400 MHz NMR spectrometer, CDCl₃ as the solvent, and a temperature of 25°C (Figure 3c). In the HSQC spectrum, the chemical shift of hydrogen (δ, ppm) is displayed by the horizontal axis, while the chemical shift of carbon (δ, ppm) is represented by the vertical axis. The connection between carbon and hydrogen is indicated by red elliptical dots. The positions of these cross-peaks allow us to methodically ascertain the carbon and hydrogen direct coupling connections in the sample, therefore confirming compound 1’s structure.

Additionally, we utilized ¹H-¹³C HMBC spectroscopy to examine the multiple bond coupling between carbon atoms and protons in compound 1 (Figure 3d). We further supported the structure of compound 1 by methodically verifying the long-range couplings between carbon atoms and protons by the analysis of the chemical shifts of the cross-peaks. We used mass spectrometry to determine its relative molecular mass following extensive NMR analysis, and the findings exhibited an actual value of 250.15 and a theoretical value of 250.16 (Figure 4). Based on this, we confirmed that compound 1 was successfully extracted from the aerial parts of Artemisia annua.

Close

Figure 4.

Mass spectrum (TOF-MS) of compound 1.

Export to PPT

3.2. The crystal structure of CP1

X-ray diffraction crystal structure analysis was conducted to identify the structure of the CP1 {Mn₅(C₂₁H₁₂NO₆)₄(H₂O)₄}. The findings suggested that CP1 was in the tripartite crystal system with the space group R-3. The asymmetric and molecular units of the complexes have been presented in Figures 5(a,b), separately. In its structural unit, three Mn²⁺ sites exhibited two types of coordinated modes, the Mn(1) site displayed five-coordinated by five O atoms from five different C₂₁H₁₂NO₆ ligands, while Mn(2) and Mn(3) were six-coordinated, different, Mn(2) site is coordinated through six O atoms come from six distinct C₂₁H₁₂NO₆ ligands to lead to a {MnO6} octahedral geometry. The Mn(3) site is linked by six O atoms derived from the neighboring four C₂₁H₁₂NO₆ ligands and two water molecules to result in a {MnO₄(H₂O)₂} octahedral geometry. Two Mn(3) cations were connected with one Mn(2) cation through bridging O atoms, forming a three-nuclear trimer {Mn3} cluster. Within the {Mn3} cluster, the distance of Mn···Mn is 3.631 Å. The Mn-O bond spacings range from 2.043(3) to 2.281(2) Å. The above lengths of bond are within the normal range. Each {Mn3} trimer cluster is linked to six C₂₁H₁₂NO₆ ligands, and each Mn(1) cation is connected with four C₂₁H₁₂NO₆ ligands. With this structure, a 3D network can be produced. Furthermore, the packing structure of CP1 revealed the presence of the C–H···O and O–H···O hydrogen bonds. Table 2 provides comprehensive details on the hydrogen bonds of 1. The hydrogen bonding details of CP1 have been given in Table 2. Figure 5(c) displayed the crystal stacking structure of CP1 along the c-axis.

Close

Figure 5.

(a) CP1’s asymmetric unit structural unit; (b) CP1’s molecular structural unit; (c) CP1’s crystal packing structure along c axis.

Export to PPT

3.3. Structural characterization of PEI-FA@CP1@1

To comprehensively evaluate the structural modifications of PEI-FA@CP1@1, the study also characterized the structure of CP1@1 under the same conditions. Figure 6(a) displays that the FTIR spectrum of the PEI-FA@CP1@1 system exhibits several prominent absorption peaks. Notably, amine groups are indicated by the broad peak at 3371 cm^-1, whereas the C-H stretching vibrations of alkyl chains are represented by the peaks at 2938 and 2893 cm^-1. The C=O stretching vibrations of carboxylic acid or amide groups are responsible for the absorption at 1613 cm^-1, whereas the C-N stretching and N-H bending vibrations of amide groups are responsible for the peak at 1511 cm^-1. Furthermore, carboxylate groups are linked to the peak at 1450 cm^-1, whereas C-O stretching vibrations that are characteristic of esters are responsible for the peak at 1033 cm^-1. These spectral characteristics reveal the material’s organic functional group structure in addition to confirming the material’s effective synthesis, which is crucial for enhancing the biocompatibility and solubility of the drug, as well as improving its targeting capabilities. Figure 6(b) presents the scanning electron microscope (SEM) image of the PEI-FA@CP1@1 system, which reveals a porous surface structure. The surface is rough and covered with fine particles or protrusions. The edges of the pores are irregular, and additional, smaller pores are scattered in the surrounding areas. The average pore size is approximately 400 nm. This porous morphology is advantageous for applications in materials science or biomedical fields, as it can enhance the surface area and potentially improve the drug delivery performance of the system. The dynamic light scattering (DLS) measurement further supports the uniformity of particle size, with a sharp peak at approximately 100 nm in the size distribution chart, consistent with the SEM observations, indicating excellent monodispersity and uniform size (Figure 6c). The TGA results, as shown in Figure 6(d), reflect the thermal stability of the sample from ambient temperature to 800°C. The sample begins to lose significant weight at about 300°C, suggesting decomposition of organic components. When the temperature exceeds 600°C, the weight remains relatively stable, indicating that the inorganic framework (such as Mn-based MOF) has good thermal stability. These data provide crucial physicochemical foundations for the potential applications of the system, confirming its suitability and reliability as a drug carrier.

Close

Figure 6.

Comprehensive characterization of PEI-FA@CP1@1. (a) FTIR; (b) SEM; (c) DLS; (d) TGA.

Export to PPT

3.4. In vitro biological evaluation of PEI-FA

PEI-FA, as an advanced carrier nanomaterial, must have its safety comprehensively evaluated before clinical application. We particularly focused on its cytotoxicity, which is a crucial indicator for assessing the safety of PEI-FA composite materials. In this study, we thoroughly examined the toxic effects of PEI-FA@CP1 and unencapsulated CP1 on HFL-1 cells. Employing concentration ranges from 0 to 20 µg/mL, we evaluated the biocompatibility of both materials. The results, as shown in Figure 7(a), indicate that cells treated with PEI-FA@CP1 consistently exhibited significantly higher viability than those treated with CP1 alone. Notably, at higher concentrations (10 and 20 µg/mL), cells treated with PEI-FA@CP1 exhibited considerably higher survival rates (84% and 81%, respectively) than those treated with CP1 only (65% and 56%, respectively), highlighting the effectiveness of PEI-FA encapsulation in significantly reducing cytotoxicity associated with the core material. Moreover, the consistently higher cell viability across all concentrations suggests that the PEI-FA@CP1 composite could allow for higher therapeutic dosages without corresponding increases in toxicity, crucial for effective cancer treatment. The stability of the composite and its interaction with cellular systems might be influenced by its controlled release properties, facilitating sustained drug action at target sites while minimizing systemic side effects. Overall, the cell viability results provide strong evidence that the PEI-FA@CP1 composite not only effectively reduces the cytotoxicity of the core material CP1 but also significantly enhances its biocompatibility; hence, it is expected to be developed for applications of targeted drug delivery. This enhanced biocompatibility, along with the potential for targeted delivery, makes the PEI-FA@CP1 composite an optimal choice for therapeutic strategies. Additionally, the drug release rate was investigated under different pH conditions (pH = 6.5, 7, and 7.4), as shown in Figure 7(b). The drug release rate over time varies significantly with pH, showing a marked increase in release as pH decreases. The highest release rates were observed under acidic conditions (pH=6.5), reaching approximately 42% within the first 2 h and approaching 85% at 12 hrs. At neutral pH (pH = 7), the release rate is moderate, with around 23% released at 2 h and 61% at 12 h. The release rate is lowest under slightly alkaline circumstances (pH = 7.4), reaching only around 8% at 2 h and 20% at 12 hrs. During the first several h, the release increases rapidly on all curves, followed by a plateau, particularly for pH = 6.5, where the rate stabilizes after 8 h. These results suggest that the drug releases more effectively in lower pH environments, making it potentially suitable for targeted release in acidic settings, such as tumor microenvironments or acidic pathological sites.

Close

Figure 7.

(a) Evaluation of the cytotoxicity of PEI-FA@CP1 and CP1. (b) The effect of pH on the release rate of compound 1.

Export to PPT

3.5. PEI-FA@CP1@1 inhibited proliferation of NSCLC cells

To assess whether PEI-FA@CP1@1 could effectively inhibit NSCLC cell proliferation, the H1975 cell line was treated with the nanoformulation, and cell viability was measured using the CCK-8 assay at 24 h and 48 h. As shown in Figure 8(a), PEI-FA@CP1@1 significantly suppressed cell viability at both time points compared to the untreated control group (P < 0.05). Given that compound 1 is known to exert pro-apoptotic effects, we next evaluated the expression of the pro-apoptotic gene Bax using quantitative real-time PCR (qPCR) (Figure 8b). Importantly, in response to the reviewer’s suggestion, we conducted a comparative analysis of free compound 1, non-targeted CP1@1, and targeted PEI-FA@CP1@1 to determine the added benefit of encapsulation and surface modification. As presented in Figure 8(c), treatment with CP1@1 resulted in a higher Bax mRNA expression level compared to free compound 1, indicating that encapsulation enhanced the bioactivity of the drug. Furthermore, PEI-FA@CP1@1 elicited the highest Bax expression, suggesting that the addition of FA-modified PEI not only improved cellular uptake via folate receptor-mediated endocytosis but also significantly enhanced pro-apoptotic activity in H1975 cells. These results demonstrate that the encapsulation of compound 1 into CP1, and further functionalization with PEI-FA, confers substantial advantages in terms of pro-apoptotic efficacy and therapeutic potential against NSCLC cells.

Close

Figure 8.

(a) The impact of PEI-FA@CP1@1 on H1975 cell proliferation. (b) The impact of PEI-FA@CP1@1 on the Bax mRNA level. (c) Comparative analysis of Bax mRNA levels in H1975 cells treated with free compound 1, non-targeted CP1@1, and folate-modified PEI-FA@CP1@1. *, P < 0.05 in contrast to the control group.

Export to PPT

3.6. Molecular docking

Machine learning was applied to screen additional compounds predicted to have enhanced anticancer effects. Although experimental validation is needed to confirm these predictions, the machine learning approach demonstrates its value by efficiently identifying promising candidates for further testing, thereby accelerating the discovery of potentially more effective anticancer agents.

The molecular docking simulation was carried out in order to determine the molecular origin of the anticancer effects of the novel designed compounds presented and demonstrated in the experimental part above. The docking pose between the protein 4JPS and the compound with the lowest binding energy (-8.03 kcal/mol) has been shown in Figure 9. A hydrogen bond was developed between the given compound and residue GLN-189 with a hydrogen bond length of 2.30 Å. Moreover, the distance between donor and acceptor was 2.96 Å, and the donor-H-acceptor angle was 122.42°. Moreover, there were multiple hydrophobic interactions surrounding the given compound, and the residues were LYS-184, LEU-185, ILE-190, and PRO-283.

Close

Figure 9.

The docking pose between the protein 4JPS and the compound with the lowest binding energy (-8.03 kcal/mol), a hydrogen bond was created between the residue GLN-189 and the given compound, the length of the hydrogen bond was 2.30 Å.

Export to PPT

Although the given compound was seen to have a relatively rigid body, the limited polar functional groups could still provide hydrogen bond acceptor and therefore exhibit anti-cancer effect. Based on such results, the generative machine learning simulation was conducted using the given compound as the input. Up to 8000 evaluations were performed, and each evaluation contained 16 modifications (addition and removal of bond or atom). The binding energy of those 8000 evaluations have been illustrated in Figure 10. It was found that the overall trend of the binding energy decreased as the generative machine learning proceeded, indicating the direction of the generative model was correct. Further, the binding energy was seen to stabilize after evaluation reached 5000.

Close

Figure 10.

The binding energy of 8000 evaluations with respect to the protein 4JPS.

Export to PPT

The SA score, which was used as the second criterion for the generative machine learning, has been presented in Figure 11. It was displayed that the SA score was narrowly distributed between 0.2 and 0.6; only limited evaluations were found outside such a range. Different from the binding energy that was shown above, the SA score in the current study was less sensitive to the generative machine learning process. This could be rationalized by the fact that the given compound was relatively rigid, whose an SA score was 0.50.

Close

Figure 11.

The SA score of 8000 evaluations, as a benchmark, the SA score for the given compound was 0.50.

Export to PPT

Further, the third generative machine learning reward, QED score, was displayed in Figure 12. Similar to the trend that was seen for the binding energy, the QED score increased as the generative machine learning proceeded and stabilized after the evaluations reached 5000. As a benchmark, the QED score for the given compound was 0.49; in contrast, the QED scores for the molecules that were generated after 5000 evaluations were aligned between 0.7 and 0.9, which are much higher than the given compound, again suggesting a successful implemented generative machine learning.

Close

Figure 12.

The QED score of 8000 evaluations, as a benchmark, the SA score for the given compound was 0.49.

Export to PPT

To further examine the results from generative machine learning, three representative structures were chosen on the basis of the binding energy, SA, and QED thresholds, where the values were set to -9.2 kcal/mol, 0.48, and 0.76, respectively. It could be seen that, besides the SA score, the thresholds for binding energy and QED score were much higher than those of the given compound, which were recorded as -7.4 kcal/mol and 0.49 (it is essential to note that the binding energy shown here for the specific compound was evaluated by Qvina for making a consistent comparison to the 8000 evaluations). In contrast, the threshold for the SA score was a little bit lower than the given compound; this was because the range of the SA score was narrow. Based on such threshold limitations, there were 25.1%, 20.0% and 24.1% of the evaluations in each criterion exhibiting better scores. Among these evaluations, three structures were selected (as listed in Figure 13), and their binding energies were -9.4, -10.0, and -9.5 kcal/mol; their SA scores were 0.53, 0.48, and 0.48; and their QED scores were 0.89, 0.86, and 0.76.

Close

Figure 13.

Three structures that were selected from 8000 evaluations, their binding energies were -9.4 (a), -10.0 (b) and -9.5 (c) kcal/mol, their SA scores were 0.53, 0.48, and 0.48, and their QED scores were 0.89, 0.86, and 0.76.

Export to PPT

Considering the generative machine learning only considered one binding mode per evaluation through Qvina, the binding pose between the protein 4JPS and chosen structures was re-evaluated through molecular docking simulations by probing 50 possible poses. The binding poses with the lowest binding energies for each of the select molecules were shown in Figure 14; the corresponding binding energies were -8.16, -6.20, and -8.02 kcal/mol. All three binding poses showed only one hydrogen bond, and the distances were 2.8 (LYS-924), 1.8 (ASP-883), and 2.1 (SER-275) Å. Such results indicated that the molecules from the generative machine learning exhibited a better anti-cancer effect at a larger scale than the given compound.

Close

Figure 14.

Binding poses for the selected structures with protein 4JPS, their binding energies were (a) -8.16, (b) -6.20, and (c) -8.02 kcal/mol, respectively.

Export to PPT