Design, Synthesis, Anticancer Screening, and Virtual Analysis of New 7-Sulfonyldiazepane- and 7-Sulfonylpiperazine-Substituted Oxazolo[4,5-d]pyrimidines

Abstract

The design of novel anticancer agents has been advanced by applying machine learning techniques, leading to several classification and regression structure-activity relationship models specifically tailored for the human glioblastoma SNB-75. Using these models, a virtual library was screened to identify compounds with potential activity against the SNB-75 cell line. A series of new 7-sulfonyldiazepane- and 7-sulfonylpiperazine-substituted oxazolo[4,5-d]pyrimidines were synthesized and evaluated for their antitumor activity against the NCI-60 cell lines. 2-(4-Methylphenyl)-5-phenyl-7-[4-(phenylsulfonyl)piperazin-1-yl][1,3]oxazolo[4,5-d]pyrimidine demonstrated cytotoxic activity and a high level of selectivity to CNS SNB-75, while 2-(4-methylphenyl)-7-[4-(methylsulfonyl)-1,4-diazepan-1-yl]-5-phenyl[1,3]oxazolo[4,5-d]pyrimidine showed very high activity against the non-small lung cancer NCI-H460 cell line. The compounds met druglike criteria according to ADMET analysis. Molecular docking investigations of synthesized compounds demonstrated complex formation with carbonic anhydrase XII as a potential mechanism. This makes these compounds attractive for further study as agents against non-small cell lung cancer and CNS cancer.

 

Scroll to top

Key words

oxazolo[4,5-d]pyrimidines - QSAR - anticancer screening - COMPARE correlation - docking

Heterocyclic compounds, particularly nitrogen- and oxygen-containing ones, play a crucial role in developing novel anticancer therapies. In recent years, most small-molecule drugs approved by the Food and Drug Administration (FDA) for treating both hematological and solid tumors have included nitrogen-containing heterocycles. Among this category, the oxazolo[4,5-d]pyrimidine framework stands out due to its fused heterocyclic structure, resembling purines, with the imidazole ring replaced by an oxazole ring. Purine analogues, classified as antimetabolites, hold particular significance in oncology, exemplified by FDA-approved medications such as azathioprine and thioguanine.[1] Purine serves as the primary structural component of purine bases, integral to DNA or coenzymes involved in nucleic acid synthesis. Many purine antagonists (e.g., mercaptopurine, thioguanine, azathioprine, or fludarabine) are used as antimetabolite drugs in cancer therapy.

Oxazolo[4,5-d]pyrimidines, in contrast to oxazolo[5,4-d]pyrimidines, are a poorly studied class of compounds due to limited access to this structure, which has led to a correspondingly limited number of biological studies.[2] The paper cited above presents data on the biological activity of oxazolo[4,5-d]pyrimidines up to 2017, showing that these derivatives have not been tested for antitumor activity. In the following years, only one article was published on this topic.[3] In this work, we present the results of the antibreast cancer potency of oxazolo[4,5-d]pyrimidine template functionalized at the 7-position with diazepane or piperazine and at the 2- and 5-positions with aromatic substituents. All synthesized compounds showed activity against the tested breast cancer cell lines (MCF7, MDA-MB-231, HS 578T, BT-549, T-47D, MDA-MB-468) in the range of micromolar concentrations. Among them, 5-phenyl-7-piperazin-1-yl-2-p-tolyl-oxazolo[4,5-d]pyrimidine exhibited the highest antitumor potency (GI₅₀ = 0.8 ± 0.33, TGI = 2.04 ± 0.56, and LC₅₀ = 5.10 ± 0.67 μM).

In addition, Karatas et al. evaluated the anticancer activity of oxazolo[4,5-d]pyrimidine bioisosters in which the pyrimidine is replaced by pyridine, which showed inhibition of human topoisomerase IIα with a potency superior to that of the anticancer drug etoposide. However, even the most active compound inhibiting hTopo IIα showed no cytotoxicity on the cancer cell lines HeLa (cervical adenocarcinoma), WiDR (colon adenocarcinoma, HT-29 derivative), A549 (lung adenocarcinoma), and MCF7 (breast carcinoma).[4]

Sulfonyl hybrids are known to have a broad spectrum of anticancer activity, low tumor resistance, and minimal side effects, making them promising molecules for cancer treatment.[5] [6] [7]

This served as the basis for functionalizing several compounds with sulfonyl substituents that showed antibreast cancer potency.

In silico methods can significantly decrease the time and effort required to identify new chemicals for anticancer drug development. By utilizing computer modeling, researchers can predict how molecules behave and interact with specific targets, enabling a more efficient screening process for potential drug candidates and helping to prioritize the most promising options.

The objective of our research, aligned with the priority tasks of designing new biologically active low-molecular-weight chemical compounds, was to investigate new oxazolo[4,5-d]pyrimidine derivatives using both in silico and in vitro methods as potential anticancer agents. Here, we present the results of the anticancer activity of 2,5-diaryl-7-piperazin-1-yl- (or -7-diazepan-1-yl)oxazolo[4,5-d]pyrimidines, in which electron-withdrawing sulfonyl groups are attached to the piperazine and diazepane cores.

Classification SAR Models

At the initial analysis stage, all molecules were processed using OCHEM software. The two-dimensional coordinates of the atoms were recalculated, and counterions and salts were removed from the molecular structures. In addition, the molecules were neutralized and mesomerized.[8] The initial dataset, consisting of 3248 different chemical compounds, was randomly divided into two subsets: a training set containing 2598 samples and a test set including 650 samples. Classification SAR models were developed according to the detailed protocols outlined in the experimental part.

As detailed in Figure [1], the classification models developed using Trans-CNF, ASNN, and RF MLT showed the best performance. For this analysis, the best-performing models for the ASNN and RF methods included the E-state, ALOGPS, and CDK2 descriptors.

Additional statistical parameters and performance metrics for each individual model are presented in Figure S1 in the Supporting Information.

Figure 1 Parameters of the classification SAR models. BA – balanced accuracy characterizes an effective measure of the reliability of the model; AUC – area under the curve serves as a valuable indicator of the overall efficiency of the model.

A consensus model was developed by calculating the simple average of the different tree models used in the study. To assess the reliability of this consensus model, its predictions were used to assess the applicability domain, ensuring that the predictions of the model remained relevant across different scenarios. Notably, all individual models showed comparable performance metrics, indicating high reliability. In terms of sensitivity, specificity, and balanced accuracy (BA), each model demonstrated effective performance, with balanced accuracy for both training and test sets impressively ranging from 93% to 95%. This high level of accuracy underlines the robustness of the models and their potential for practical application in real-world situations.

 

Scroll to top

Regression SAR Models

Regression models were developed similarly to classification models according to the protocols outlined in the Experimental part. To create the external validation set, the original dataset of 2340 compounds was partitioned into the training set of 1837 compounds, representing approximately 80% of the total dataset, and the test set of 465 compounds, representing 20%. The Trans-CNF and Trans-CNN machine learning approaches demonstrated the most excellent effectiveness in generating QSAR models. A consensus model, representing the average of the two models, was used to provide a comprehensive quantitative assessment of potential anticancer inhibitors, as detailed in the following section. The predictions generated by the consensus model were also utilized to evaluate its applicability range.[9] The q² values ranged from 0.71 to 0.72 for the training sets and between 0.72 and 0.73 for the test sets. Table [1] and the Supporting Information (Figure S2) present additional statistical parameters relevant to the models.

Table 1 Statistical Coefficients of the QSAR Models Were Calculated Using the Three MLMs

Entry	Method	Set	Size	R2	q2	RMSEb
1	Trans-CNF	training	1837	0.72 ± 0.01	0.72 ± 0.01	0.45 ± 0.01
		test	465	0.72 ± 0.02	0.72 ± 0.02	0.44 ± 0.01
2	Trans-CNN	training	1837	0.71 ± 0.01	0.71 ± 0.01	0.46 ± 0.01
		test	465	0.72 ± 0.02	0.71 ± 0.01	0.44 ± 0.01
3	consensus modela	training	1837	0.71 ± 0.01	0.72 ± 0.01	0.44 ± 0.01
		test	465	0.73 ± 0.02	0.73 ± 0.02	0.43 ± 0.01

^a The consensus model was a simple average of the Trans-CNF and Trans-CNN models.

^b RMSE is the root mean square error. R² and q² are the squared linear correlation and coefficient of determination.

The regression line shown in Figure [2] illustrates the predicted values generated by the consensus QSAR model specifically tailored for glioblastoma inhibitors. Analysis of the performance of this model shows that the most of its predictions are in close agreement with the experimental values, with no more than 1 log unit discrepancies.

Figure 2 Plots of experimental versus predicted values for the consensus QSAR model; GI₅₀ is the concentration that produces a 50% reduction in net cell growth. The goodness of fit parameters are shown in Table [1].

Within the training set, only 34 compounds, representing a mere 1.85%, exhibit residuals that exceed a difference of 1 log unit between the predicted and experimental log GI₅₀ values. This indicates a relatively high level of accuracy in the predictions of the model.

In the independent test set, the reliability of the model further solidifies, with only three compounds, or approximately 0.6%, displaying residuals greater than 1 log unit but less than 1.4 log units. This minimal deviation underscores the robustness and predictive capability of the consensus QSAR model, as evidenced by the data shown in Figure [2], complemented by the detailed analysis in Table [1] and Figure S2 (see the Supporting Information). These results collectively affirm the effectiveness of the model in accurately forecasting the activity of glioblastoma inhibitors based on chemical structure.

 

Scroll to top

Prediction of Antiviral and Antitumor Activity of New Compounds

A virtual dataset was created, encompassing 22 diverse thiazolo- and oxazolo[4,5-d]pyrimidine derivatives with different substitution patterns. This dataset was meticulously generated through an in-depth analysis of existing literature[1] [10] and the valuable insights provided by seasoned synthetic organic chemists. Detailed information regarding the specific derivatives can be found in Table S1 of the Supporting Information. The eleven compounds predicted as active ones among those with the most confident predictions (> 80%) were selected for further evaluation (Table S1) by using binary classification models.

In the subsequent phase of the study, we used regression models to evaluate the efficacy of various compounds in inhibiting cell growth against the central nervous system SNB-75 cell line. The results of this analysis are summarized in Table [2]. The log(1/GI₅₀) values were predicted to range from 5.4 to 5.65, indicating potency at a concentration below 10 μM for all compounds tested. Moreover, all compounds were evaluated for their activity against the K-562 leukemia cell line using the previously published consensus model.[11] This assessment further validated that these compounds exhibit potential efficacy against this type of cancer at minimum inhibitory concentration (MIC) levels of less than 10 μM. Based on the predictions of anticancer activity, all eleven compounds were chosen for subsequent synthesis and thorough biological testing (Table [3]).

Table 2 Evaluation of Antitumor Activity of Eleven Compounds Using Two Consensus Models

Compound	CNS cancer SNB-75 cell line			Leukemia K-562 cell line
	log(1/GI50)a (M)	Consensus stdc	AD	log(1/MIC)b (mol/L)	Consensus stdc	ADd
1	5.57	0.01	true	5.40	0.49	true
2	5.61	0.03	true	5.39	0.55	true
3	5.66	0.05	true	5.52	0.39	true
4	5.64	0.04	true	5.39	0.35	true
5	5.64	0.04	true	5.39	0.36	true
6	5.54	0.14	true	5.57	0.81	false
7	5.44	0.03	true	5.44	0.58	true
8	5.42	0.01	true	5.43	0.68	false
9	5.54	0.10	true	5.67	0.67	false
10	5.46	0.07	true	5.45	0.57	true
11	5.49	0.02	true	5.47	0.58	true

^a GI₅₀ is the concentration for 50% of the maximal inhibition of cell proliferation.

^b MIC – minimum inhibitory concentration.

^c Consensus std – the standard deviation of the predictions obtained from an ensemble of models.

^d AD – applicability domain.

Table 3 7-Sulfonylpiperazine- and 7-Sulfonyl-1,4-diazepane-Substituted 2,5-Diaryloxazolo[4,5-d]pyrimidines 1–11

Compound	NSC code	Structure	Name
1	811841		2,5-diphenyl-7-[4-(phenylsulfonyl)-1,4-diazepan-1-yl][1,3]oxazolo[4,5-d]pyrimidine
2	811842		7-[4-(4-methylphenylsulfonyl)-1,4-diazepan-1-yl]-2,5-diphenyl[1,3]oxazolo[4,5-d]pyrimidine
3	811846		2-(4-methylphenyl)-7-[4-(methylsulfonyl)-1,4-diazepan-1-yl]-5-phenyl[1,3]oxazolo[4,5-d]pyrimidine
4	811855		2-(4-methylphenyl)-5-phenyl-7-[4-(phenylsulfonyl)-1,4-diazepan-1-yl][1,3]oxazolo[4,5-d]pyrimidine
5	811856		2-(4-methylphenyl)-7-[4-(4-methylphenylsulfonyl)-1,4-diazepan-1-yl]-5-phenyl[1,3]oxazolo[4,5-d]pyrimidine
6	811838		7-[4-(methylsulfonyl)piperazin-1-yl]-2,5-diphenyl[1,3]oxazolo[4,5-d]pyrimidine
7	811839		2,5-diphenyl-7-[4-(phenylsulfonyl)piperazin-1-yl][1,3]oxazolo[4,5-d]pyrimidine
8	811840		7-[4-(4-methylphenylsulfonyl)piperazin-1-yl]-2,5-diphenyl[1,3]oxazolo[4,5-d]pyrimidine
9	811843		2-(4-methylphenyl)-7-[4-(methylsulfonyl)piperazin-1-yl]-5-phenyl[1,3]oxazolo[4,5-d]pyrimidine
10	811844		2-(4-methylphenyl)-5-phenyl-7-[4-(phenylsulfonyl)piperazin-1-yl][1,3]oxazolo[4,5-d]pyrimidine
11	811845		2-(4-methylphenyl)-7-[4-(4-methylphenylsulfonyl)piperazin-1-yl]-5-phenyl[1,3]oxazolo[4,5-d]pyrimidine

 

Scroll to top

Synthesis of Oxazolo[4,5-d]pyrimidine Derivatives

A convenient and preparative approach for the synthesis of oxazolo[4,5-d]pyrimidine derivatives was developed in our department based on substituted 1,3-oxazol-5(4H)-ones Ia,b (Scheme [1]).[12] The interaction of equimolar amounts of oxazolones Ia,b with amidines IIa,b in tetrahydrofuran in the presence of triethylamine produces substituted imidazol-4(5H)-ones A, which, when heated in pyridine, undergo recyclization to form the corresponding oxazolo[4,5-d]pyrimidin-7-ones IIIa–c in high yields. Heating the latter in phosphorus oxychloride in the presence of an equimolar amount of N,N-dimethylaniline affords oxazolo[4,5-d]pyrimidines IVa–c, which contain a mobile chlorine atom in the pyrimidine ring, capable of substitution reaction with various N-nucleophiles, in particular, N-unsubstituted diazepanes (Va,b) or piperazines (VIa,b).

Scheme 1 Synthesis of 2,5-diaryl-7-chlorooxazolo[4,5-d]pyrimidines IVa–c

The synthesis of N-sulfonyl derivatives of oxazolo[4,5-d]pyrimidines 1–11 was carried out by the reaction of N-unsubstituted piperazine and 1,4-diazepane derivatives of oxazolo[4,5-d]pyrimidines Va,b and VIa,b with methane- or arenesulfonyl chlorides. The reaction occurs upon heating in dioxane in the presence of triethylamine (Scheme [2]). Elemental analysis, IR, ¹H, and ¹³C NMR spectra, and LC-MS spectrometry confirmed the structures of compounds 1–11.

Scheme 2 Synthesis of 7-sulfonylpiperazine- and 7-sulfonyl-1,4-diazepane-substituted 2,5-diaryloxazolo[4,5-d]pyrimidines 1–11

 

Scroll to top

Screening of Oxazolo[4,5-d]pyrimidine Derivatives

The One-Dose Assay

The complete screening data are presented in Table S1 (see the Supporting Information). When calculated for the entire panel most of the compounds tested exhibited no cell growth inhibition. Only compounds 3 and 10 demonstrated weak and very weak inhibition of the growth of the full panel of cell lines (31% and 14%, respectively), with significant heterogeneity in the growth response of the individual cell lines (Tables S1 and S2). The most sensitive cell lines to these compounds are detailed in Table [4].

Table 4 Growth Inhibitory Activity (GI) of Compounds 3 and 10 against the Most Sensitive Cell Lines (GI ≥ 70%)

Compound	Cancer subpanel				Total panel	SIa
	NSCL	Colon	CNS	Renal
3	NCI-H460 (96)a	HCT-116 (76)	SNB-75 (83)	RXF 393 (81)	31	3.1
10	–	–	SNB-75* (117)	–	14	8.4

^a SI – Selectivity index to appropriative cell line, SI = GI_cell/GI_total.[13] Ratios SI between 3 and 6 refer to moderate selectivity, ratios greater than 6 indicate high selectivity toward the corresponding cell line, while compounds not meeting either of these criteria are rated non-selective.[13]

Compound 3 demonstrated selective and very high cytostatic activity, nearly complete, against the NSCL NCI-H460 cell line, but it did not display selectivity regarding other highly sensitive cell lines. In addition, this compound moderately inhibited (50 ≥ GI < 70%) the growth of leukemia SR (69%), NSCL A549/ATCC (61%), colon HCT-116 (76%) and SW-620 (56%), CNS SF-539 (69%), melanoma MDA-MB-435 (65%), ovarian OVCAR-4 (52%), and renal ACHN (54%) cell lines. Compound 10 is the only compound that exhibited cytotoxic activity.

Of all the cell lines in the NCI panel, only the CNS SNB-75 cell line showed a cytotoxic response to compound 10, exhibiting high GI selectivity relative to the GI of the total panel (SI = 8.4). Among the remaining cell lines in the total panel, compound 10 moderately inhibited the growth of only the breast cancer HS 578T (57%) cell line. This indirectly suggests a unique target specificity of this compound concerning the molecular target involved in its cytotoxic activity. Examples of such specificity include the FDA-approved anticancer drugs with kinase activity, dasatinib, which is cytotoxic at HICON = 10 μM in the melanoma subpanel only against the LOX-IMVI cell line, and imatinib, which shows high growth inhibitory activity in the overall panel only against the leukemia K-562 cell line, the only line harboring the BCR-ABL translocation.[14] The remaining compounds exhibited very weak growth inhibitory activity against individual cell lines, not exceeding 30%.

An attractive property of compounds 3 and 10 is their selectivity, suggesting that they may also be preserved for normal cells of the exact origin due to their genotypic similarity. This, in turn, indicates a reduced likelihood of serious side effects during their therapeutic use.

 

Scroll to top

COMPARE Correlation

The COMPARE test is a common method used to determine the likelihood that the mean bioactivity graph of a test compound will be similar to that of a reference compound with a known molecular mechanism of action. If the test and reference compounds have comparable mean graphs for inhibition of tumor cell growth, this may indicate the presence of similar molecular mechanisms of action. The degree of similarity can be quantified by calculating the Pearson correlation coefficient, r. A COMPARE analysis (Figure [3]) revealed that compound 10 exhibited only moderate correlation (0.50 ≤ r < 0.70) with a mere three standard drugs for each parameter characterizing antiproliferative activity (with VP-16 (etoposide), A-TGDR, and nitroestrone – for GI₅₀ vector, and with didemnin B, nitroestrone, and ftorafur – for TGI vector). No standard drugs were found that were even moderately close to compound 10 in cytotoxicity (r ≤ 0.50), indirectly indicating that this derivative has a unique cytotoxic action mechanism. Compound 3 exhibited a moderate correlation with aclacinomycin A for the GI₅₀ vector. Furthermore, the lack of similarity (no records found for p ≥ 0.3) between the TGI vectors of the reference drugs and compound 3 suggests that the mechanisms of antiproliferative action inherent in the reference compounds do not include their cytostatic activity.

Figure 3 Standard drug vectors that showed the highest correlations with compounds 3 and 10 according to the one-dose assay data

 

Scroll to top

ADMET Analysis

The use of ADMET filters is a common practice that aids in identifying small molecules for drug discovery candidates, helping prioritize molecules for further studies. The selection of molecules relies on passing the chosen filters, depending on the intended purpose of the synthesized compounds. This approach minimizes the screening of drug candidates in clinical trials due to undesirable pharmacological properties.[15] The classification of drug similarity can be carried out according to two distinct criteria. Firstly, drug similarity can be defined as general drug similarity, which includes drugs used to treat various diseases. Secondly, it can be classified as specific drug similarity based on specific diseases, molecular targets, and particular classes of compounds, among other factors.

 

Scroll to top

Physicochemical Properties

These properties are crucial in ADMET analysis, as they are used to predict various parameters describing the biomedical characteristics of the molecules in question. To assess the physicochemical properties and overall drug similarity of compounds 3 and 10, data calculated by the Drug-Like Soft filter developed by FAFDrugs4 was employed.[16a] For the virtual assessment of the physicochemical properties related to drug similarity, the FAFDrugs4 filter used a chemical space encompassing up to 90% of the oral drugs listed by the FDA to propose the filter thresholds. Given that both compounds showed inhibitory activity against central nervous system (CNS) cell lines, a specific drug identity filter for CNS-penetrating compounds, proposed by Jeffrey et al., was utilized.[16b] The physicochemical parameters of compounds 3 and 10, as predicted by ADMETlab3.0, are presented in Table [5], along with the threshold values of druglikeness physicochemical properties adopted by the aforementioned filters.

Table 5 The Parameters of Physicochemical Properties of Compounds 3 and 10, and Threshold Values Druglikeness According to Drug-Like Soft and CNS Rules

Descriptora	Compound		Drug-Like Soft rule	CNS rule
	3	10
MW	463	511	100–600	135–582
logP	3.6	4.2	–3 to 6	–0.2 to 6.1
logD	4.6	5.4	–	–
nHA	8.0	8.0	≤ 12	≤ 5
nHD	0.0	0.0	≤ 7	≤ 3
tPSA	92	92	≤ 180	3–118
nRot	4.0	5.0	≤ 11
nRig	31	36	≤ 30
nRing	5.0	6.0	≤ 6
MaxRing	9.0	9.0	≤ 18
nCarb	24	28	3–35
nHet	9.0	9.0	1–15
fChar	0.0	0.0	≤ 4

^a Descriptors: MW – molecular weight, LogP – octanol/water distribution coefficient, nHA – number of H-bond acceptors, nHD – number of H-bond donors, tPSA – polar surface area, nRot – number of rotatable bonds, nRig – number of rigid bonds, nRing – number of rings, MaxRing – number of atoms in the biggest ring, nCarb – number of carbon atoms, nHet – number of heteroatoms, fChar – formal charge.

According to the Drug-Like Soft rule, the compounds were above the cut-offs for two properties (see Table [5]). This suggests that there may be problems with bioavailability. However, the compounds did not exceed the threshold values set out in the CNS rule, with only one violation of the filtering conditions. These physicochemical properties of the compounds serve as descriptors for various filters, which use different databases as drug training sets to evaluate the oral bioavailability and safety profile of drug candidates. ADMETlab3.0 predicts that both compounds meet the Pfizer rule (logP > 3, TPSA > 75). Compound 3 also satisfies the Lipinski rule (MW ≤ 500; logP ≤ 5; nHA ≤ 10) and the Golden Triangle rule (MW ≤ 500, –2 ≤ logD ≤ 5). However, when the descriptor cut-offs derived from the database used by FAFDrugs4 are applied to the drug similarity filters, these compounds fulfill most of the binary filters (Lipinski, Ghose, Veber, Egan, Muegge, and GlaxoSmithKline rules).

 

Scroll to top

Pharmacokinetic Properties

ADMET analysis enables the prediction of pharmacokinetic profiles of the compounds, which is crucial for assessing their pharmacodynamic activity. Table [6] presents the predicted pharmacokinetic properties of compounds 3 and 10.

As the predictive confidence for compound 3 is situated within a probability range that precludes drawing of a definitive conclusion, an additional analysis of the relevant parameters was performed using the Deep-PK platform harnessing deep learning algorithm (see ref. 50).

Table 6 Predicted Pharmacokinetic Profile of Compounds 3 and 10

Parametera	Compound		Decision
	3	10	3	10
Absorption
Caco-2 Permeability, Papp (cm/s)	20.8 10–6	16.7 10–6	high (Papp > 8 × 10–6)
MDCK Permeability, Papp (cm/s)	13.1 10–6	21.8 10–6	medium (2–20 × 10–6) high (Papp > 20 × 10–6)
PAMPA, logPeff	> 2.5	> 2.5	high
HIA%	> 30	> 30	absorbable
F50%	≥ 50	≥ 50	bioavailable
Distributionsb
PPB	≥ 98%		high
Fu (%)	1.353	0.729	low
VDss (L/kg)	0.125	0.14	optimal (0.04–20 L/kg)
BBB permeant	0.515	0.91	FA	high
Pgp substrate	0.41 (0.299c)	0.05	FA (noc)	no
Pgp inhibitor	0.999	1.0	yes
MRP1 inhibitor	0.999	0.993
BCRP inhibitor	0.007	0.0	no
Metabolismb
CYP1A2: inhibitor substrate	0.76 0.07	0.03 0.0	yes no	no
CYP2C19: inhibitor substrate	0.07 0.22	0.94 0.03	no	yes no
CYP2C9: inhibitor substrate	0.58 (0.78c) 0.46	0.71 0.99	FA (noc) FA	yes
CYP2D6: inhibitor substrate	0.002 0.14	0.0 0.03	no
CYP3A4: inhibitor substrate	0.007 0.43 (0.86c)	0.75 0.93	no FA (yesc)	yes
CYP2B6: inhibitor substrate	0.09 0.14	0.1 0.0	no
CYP2C8 inhibitor	0.99	1.0	yes
Excretion
CL (mL/min/kg)	4.1	5.5	low	moderate
Halflife of Drug (h)	0.50	0.29	ultra-short halflife

^a Parameters: Papp – apparent permeability coefficient, P_eff – resulting effective permeability, HIA – human intestinal absorption. Molecules with HIA > 30% were classified as absorbable. PPB – plasma protein binding, optimal: < 90%. F_u – the fraction unbound in plasma, low: < 5%. VDss – volume distribution, optimal range: 0.04–20 L/kg. F50% – human oral bioavailability ≥ 50%. CL – plasma clearance, moderate: 5–15 mL/min/kg; low: < 5 mL/min/kg. FA – further analysis required. Additional evaluation is recommended not to rely on predictive confidence with probabilities of 0.3 > P < 0.7.[17] [18]

^b Probabilistic assessment (0 ≥ P ≤ 1) if units of measurement are not specified.

^c Deep-PK data.

The results do not suggest its substrate affiliation with Pgp (P = 0.299) and allow for its classification as a CYP2C9 inhibitor (P = 0.78) and CYP3A4 substrate (P = 0.86), leaving uncertainty regarding its substrate affiliation with CYP2C9 (P = 0.35) and BBB permeant (P = 0.59). Cytochrome P450 (CYP) enzymes are the primary enzymes implicated in phase II drug metabolism, responsible for approximately 75% of human oxidative drug metabolism. Beyond their role in elimination, CYP enzymes exert substantially influence parameters such as safety, bioavailability, and the emergence of drug resistance.[19] Considering the results of the Deep-PK, it was deduced that compound 3 would demonstrate substrate specificity for CYP3A4 alone, while compound 10 would also be a substrate for CYP2C9. This suggests the potential for these cytochromes to influence the clearance and resistance of tumors to the compounds in question. Both compounds exhibited high predictive confidence for inhibitory activity against CYP2C8 and CYP2C9. Furthermore, it was predicted that compound 3 would demonstrate a high probability of inhibiting CYP1A2, and that compound 10 would show a high probability of inhibiting CYP2C19 and CYP3A4. This suggests the possibility of test compounds interfering with the pharmacokinetics of drugs metabolized by these cytochromes.

The rate of drug transport across the blood-brain barrier (BBB) obtained in silico models is used to identify promising candidates for developing drugs for the central nervous system. This function, along with passive permeability, also includes reverse active transport carried out by ATP-dependent transporters: P-glycoprotein (Pgp: ABCB1), breast-cancer-associated protein (BCRP: ABCG2) and multidrug resistance-associated protein 1 (MRP1: ABCC1). These are highly expressed at the BBB and are the main transporters of ABC efflux in the BBB.[20] These proteins prevent anticancer drugs from entering the brain environment via their reverse efflux into the bloodstream. Therefore, CNS drug candidates must demonstrate, along with high passive permeability, a lack of substrate specificity for these transporters, which both compounds demonstrated. In addition, it is well established that Pgp inhibitors do not generally interact with nucleotide-binding domains.[21] The predicted affiliation of the compounds with Pgp inhibitors enhances confidence that Pgp does not transport them. It is also suggested that the compounds bind to an allosteric site on the enzyme. Both compounds are predicted to have a high probability of Pgp and MRP1 inhibition and, conversely, a very low probability of BCRP inhibition. Consequently, these compounds may influence the distribution of MRP1 substrates, thereby modifying the efficacy of concomitant therapeutics. It is important to note that in situ permeability can be influenced not only by the rate of penetration through the BBB but also by affinity of the drug to brain tissue. As more lipophilic molecules exhibit higher tissue binding affinity, the absorption effect of tissue binding becomes significant. Conversely, with a decrease in lipophilicity, passive membrane permeability exerts a greater influence on the absorption of the drug into the CNS.[22] Because the predicted lipophilicity for both compounds approaches the given upper threshold, their lipophilicity significantly affects the BBB passive permeability.

The predicted high PPB and low F_u of the compounds suggest reduced availability of the free compound, which could compromise their efficacy. Nevertheless, the predicted high absorption capacity of these molecules in the human intestine, as well as their good permeability to the CNS and sensitivity to them by CNS cancer cell lines (SNB-75, SNB-19, and SF-539) demonstrated in vitro, clearly indicate the desirability of their further study in a five dose assay.

 

Scroll to top

Medicinal Chemistry

Identifying compounds with undesirable properties, including interfering molecules, is an important task in predicting drug similarity. These molecules exhibit a strong response that is independent of the target protein, are chemically reactive, or can interact specifically with multiple targets, which are factors responsible for poor pharmacokinetic properties. The main characteristics of these parameters for synthesized compounds are presented in Table [7].

Table 7 Filtering Results of Tested Compounds for Undesirable Properties Predicted by ADMETLab3.0

Filters	Compound		Decision
	3	10
Brenka	0 alert		acceptable
PAINSb
BMSc
Promiscuous compounds	0.02	0.01
Reactive compounds	0.005	0.005
MCE-18d	65	72	sufficient novelty
Synthetic accessibility	2		acceptable
GASAe	0.0

^a The Brenka rule filters out compounds with poor pharmacokinetics.

^b PAINS – criterion for excluding compounds that interfere with the analysis.

^c BMS - the Bristol-Myers Squibb rule is used to filter undesirable reactive compounds that could cause serious toxicities.

^d MCE-18 (medicinal chemistry evolution) – allows one to effectively evaluate molecules for novelty in terms of their total sp³ complexity. MCE-18 ≥ 45 is considered a suitable value; this measure can effectively score molecules by novelty in terms of their cumulative sp³ complexity synthetic accessibility - a synthetic availability index ≤ 6 is an indication that the compound is readily synthesizable.

^e GASA – synthetic accessibility score is designed to estimate the ease of synthesis of druglike molecules based on a combination of fragment contributions and a complexity penalty. Easy to synthesize; HS: Hard to synthesize; the output value represents the probability of being challenging to synthesize, ranging from 0 to 1.

The compounds passed all filters that removed agents with undesirable properties. Additionally, both parameters characterizing the availability of molecule synthesis demonstrated ease of synthesis, and MCE-18 falls within the range of values that allow the compounds to be classified as novel compounds.[23]

 

Scroll to top

Probability of Interaction with Molecular Targets

Probabilistic assessment of compound interactions with specific targets can be used in the early stages of anticancer drug discovery to predict their molecular mechanisms of action. Among the stress response pathways presented in the Tpx21 database, the ADMETlab3 platform predicts a very high probability of interaction with the antioxidant response element for both compounds (P > 0.93). Compound 3, in contrast to 10, also indicates a high probability of interaction with the nuclear estrogen receptor (P = 0.89). However, the high growth inhibitory activity demonstrated by the compound did not involve hormone-sensitive cell lines, allowing this target to be excluded. The probability of interaction with other stress pathways (ATPase family, AAA domain-containing protein, heat shock factor response element, mitochondrial membrane potential, and a tumor suppressor protein p53) and nucleic receptors (androgen, aryl hydrocarbon, aromatase, glucocorticoid, thyroid receptors, and peroxisome proliferator-activated receptor gamma) was predicted to be low for both compounds.

Among the targets involved in carcinogenesis, Swiss Target Prediction predicted only a probable interaction of compound 3 with MAP kinase p38 alpha, which is considered a potential oncogenic factor in brain cancer. Activation of this protein promotes glioblastoma initiation and proliferation.[24] It can be hypothesized that this target, in contrast to compound 10, is incorporated within the molecular mechanism of the antiproliferative effect of this compound on CNS cancer cell lines that are sensitive to it.

 

Scroll to top

Molecular Docking

Also, it is known that the surface of SNB-75 cells is characterized by high-level expression of carbonic anhydrase XII, which is associated with the growth and invasiveness of glioblastoma tumors.[25] Therefore, given the high anticancer selectivity of compounds 3 and 10 against SNB-75, molecular docking studies were conducted on the carbonic anhydrase XII active center. Molecular docking of ligands 3 and 10 was conducted into the active center of carbonic anhydrase XII. The 3D structure of carbonic anhydrase XII (PDB ID: 3MNA) was obtained from the RCSB global Protein Data Bank.[26] The molecular docking procedure was approved primarily by redocking the co-crystallized ligand-1. Next, a molecular docking study of compounds 3 and 10 was carried out. The results are displayed in Table [8] and Figure [4].

Table 8 The Description of Molecular Docking Results of the Ligands 1–11

Compounda	ΔG (kcal/mol)	Hydrogen bonds and amino acids		Electrostatic interaction and amino acids
1	–8.3	2	HIS64, ASN62	1	HIS94
2	–8.4	2	HIS64, ASN62	2	HIS94, ASN62
3	–8.9	6	THR199, THR200, HIS96, ASN67 (2), ASN62	0	–
4	–7.8	2	HIS64, PRO201	1	HIS94
5	–8.5	2	ASN62, HIS64	1	HIS94
6	–8.3	3	HIS64 (2), PRO201	0	–
7	–7.9	2	TRP5, GLN92	2	TRP5, HIS94
8	–8.0	1	ASN67	2	HIS94, HIS64
9	–8.2	2	HIS64, PRO201	1	HIS94
10	–9.7	5	HIS94, THR200 (2), HIS96, ASN67	2	HIS96, HIS64
11	–8.1	1	ASN67	2	HIS94, HIS64
liganda	–7.7	4	LEU198, THR199 (2), HIS94	1	TRP209

^a Methyl 2-[[4-chloro-6-(4-sulfamoylanilino)-1,3,5-triazin-2-yl]amino]acetate.

The redocking of methyl 2-[[4-chloro-6-(4-sulfamoylanilino)-1,3,5-triazin-2-yl]amino]acetate in the active site of human carbonic anhydrase XII shows a calculated binding energy of –7.7 kcal/mol and the RMSD value of –1.84 Å. The docking results indicate the formation of ligand-receptor complexes of all compounds. Low-active compounds 1, 2, 4–9, and 11 form ligand-enzyme complexes with calculated binding energies from –7.9 to –8.5 kcal/mol. Complex formation is accompanied by only 1–2 hydrogen bonds and 1–2 electrostatic interactions. However, compounds 3 and 10 form strong ligand-enzyme complexes with calculated binding energies of –8.9 and –9.7 kcal/mol, respectively. This complexation is accompanied by 4–6 hydrogen bonds, electrostatic and hydrophobic interactions. It should be noted that compounds 3 and 10 dock with lower energy than the co-crystallized ligand. Next, Figure [4] shows features of the complexation of most activity compounds 3 and 10 into the carbonic anhydrase XII active site.

Figure 4 Molecular docking of compounds 3 and 10 into the active center of carbonic anhydrase XII; a, b – compound 10; c, d – compound 3.

Compound 10 formed enzyme-ligand complex is stabilized: three hydrogen bonds between sulfonyl group and amino acid residues HIS94 (2.35 Å), THR200 (2.45 Å), and HIS96 (2.73 Å); one hydrogen bond among piperazine ring and THR200 (3.44 Å) and one hydrogen bond between the oxazolo[4,5-d]pyrimidine ring and amino acid ASN67 (2.52 Å). This complex is also stabilized by two electrostatic interactions: between the sulfur atom and HIS96 (5.77 Å), and between the phenyl ring and HIS64 (3.49 Å). Additionally, this complex is supported by many hydrophobic interactions with amino acids TRP5, HIS64, VAL121, VAL143, and LEU198 (3.40–4.91 Å). Besides, compound 3 forms an enzyme-ligand complex that is stabilized by two hydrogen bonds between the sulfonyl group and amino acid residues THR199 (2.56 Å), and THR200 (2.07 Å); one hydrogen bond among the diazepane ring and THR200 (3.77 Å) and three hydrogen bond between the oxazolo[4,5-d]pyrimidine ring and amino acids ASN67 (1.81 Å) and ASN62 (2.87–2.93 Å). This complex is also supported by multiple hydrophobic interactions with amino acids TYR7, HIS94, ALA65, THR200, LEU198, PRO202, PHE131, ILE91 (3.63–5.31 Å). Therefore, the molecular docking results of compounds 3 and 10 showed a probable mechanism of their antitumor action via binding to the active site of carbonic anhydrase XII, similar to other inhibitors (PDB ID: 3MNA, 7PP9). It should be emphasized that low-active compounds 1, 2, 4, 9, and 11 form hydrogen bonds exclusively between the sulfonyl groups and the nearest amino acid residues.

 

Scroll to top

 

Scroll to top

Conclusion

This study investigated the anticancer activity of novel synthesized oxazolo[4,5-d]pyrimidine derivatives. Initially, predictive classification and regression models were created using the OCHEM platform, incorporating various machine-learning techniques. The resulting quantitative structure-activity relationship models exhibited commendable stability, robustness, and predictive capability, as validated through cross-validation and an evaluation of a randomly generated test set.

The CNS SNB-75 cell line is expressed in human brain glioblastoma. Glioblastoma is the most aggressive and common type of brain tumor in the central nervous system. Therefore, the cytotoxic activity and high selectivity of compound 10 (SI = 8.4), in contrast to the properties of compound 3 (SI = 3.1), make it a more attractive candidate for further study as an anti-glioblastoma agent. Compound 3 showed selectivity and high antiproliferative activity against the NCI-H460 non-small cell lung cancer cell line. It also demonstrated high growth inhibition of the HCT-116 colon, CNS SNB-75, and RXF 393 renal cell lines, indicating the need further to investigate the potential of compound 3 as an anticancer agent. Several factors support this conclusion. Firstly, both compounds demonstrate congruence with the ADMET predicted values. Secondly, both compounds are novel and satisfy the druglikeness criteria. Thirdly, the TGI vectors and LC₅₀ of compounds 3 and 10 do not resemble those of the standard drugs. This suggests that the antiproliferative and cytotoxic mechanisms of action of the compounds are unique. Molecular docking analysis of compounds 3 and 10 demonstrated their potential anticancer mechanism of action as carbonic anhydrase XII inhibitors, suggesting prospects for developing a new class of compounds with anticancer activity based on them.

 

Scroll to top

 

Scroll to top

Datasets

The dataset I includes 2248 compounds from the ChEMBL database, focusing on PubChem bioassays related to anticancer activity against the SNB-75 cell line, a human brain tumor model.[27] This cell line is significant for research on glioblastoma, the most common and aggressive brain cancer in adults, which has an average survival of only 15 months post-diagnosis. Gliomas originate from glial cells and are the most prevalent central nervous system malignancies, making cell lines like SNB-75 vital for understanding brain tumor biology.

In the first step, the initial data were uploaded into the Instant Jchem database[28] and systematically ranked according to their Tanimoto Index (TI) values. All compounds identified as ‘active’ inhibitors were included in this analysis. Inactive compounds were carefully selected using the Kennard–Stone design methodology,[29] focusing on the first 10,000 bioassay compounds ranked in descending order of their TI scores. This approach ensured the creation of a diverse and informative subset of the data. The refined dataset was then uploaded to the OCHEM database for further processing. The preliminary evaluation phase carefully removed any internal duplicates within the dataset, resulting in a final curated set of 1608 active and 1640 inactive inhibitors, providing a robust basis for further research and analysis.

The second dataset, comprising 2302 compounds sourced from the ChEMBL database, was used for regression structure-activity relationship (SAR) modeling.[8] These compounds were meticulously curated based on their demonstrated activity against the human CNS SNB-75 cell line. Growth inhibition (GI₅₀) values were evaluated within a range of nanomolar (nM) concentrations, extending from 0.68 nM to 77000 nM. The GI₅₀ value quantitatively represents the concentration of a specific compound required to achieve a 50% reduction in net cell growth, serving as a key indicator of the compound’s efficacy in inhibiting cell proliferation. This measurement is critical for assessing the potential therapeutic effects of various substances on cell viability. To improve the analysis, the activity data were transformed into a log(1/GI₅₀) format, which allows for a more normalized and interpretable representation of the biological activity of these compounds. The dataset was then uploaded to the OCHEM database, a cheminformatics platform that supports the QSAR modeling process.[8] The OCHEM website provides public access to the chemical structures corresponding to the compounds used for QSAR modeling and a complete list of publications.[30]

 

Scroll to top

OCHEM Tools

While developing of QSAR models, we experimented with different machine-learning methods (MLM) and sets of descriptors available in OCHEM. In this study, the most successful models were developed using the OCHEM platform by applying four distinct machine learning methodologies: Transformer Convolutional Neural Network (Trans-CNN),[31] Transformer Convolutional Neural Fingerprint (Trans-CNF),[32] Associative Neural Networks (ASNN),[33] and Random Forest Regression (RFR).[34] The optimized parameter settings were used for each machine-learning method offered by the OCHEM platform.

The optimal ASNN and RF models were developed by combining E-state indices,[35] AlogPS,[36] and CDK2[37] software packages. These models leverage the unique benefits of each software; E-state indices provide molecular descriptors that capture electronic properties, while AlogPS significantly enhances predictive accuracy through its modeling of lipophilicity, represented as logP, and water solubility, denoted as logS, for various chemical substances. CDK2 is an advanced computational tool that calculates 256 different molecular descriptors. These descriptors act as quantitative representations of molecular properties and characteristics, covering a wide range of categories. Among them are topological descriptors, which illustrate the connectivity and arrangement of atoms within a molecule; geometrical descriptors, which focus on the spatial arrangement and distances between atoms; constitutional descriptors that detail the molecular composition; electronic descriptors related to electron distribution and properties; and hybrid descriptors that combine multiple types of information to provide a more comprehensive view of the behavior of a molecule. This versatility makes CDK2 a valuable resource in cheminformatics and computational chemistry, assisting researchers in analyzing and interpreting molecular structures and their implications.

The performance of the QSAR models was assessed through a fivefold cross-validation method and external validation sets.[38] To guard against inaccurate model estimations resulting from overfitting due to variable selection, OCHEM conducts multiple repetitions of all stages involved in model development within each validation fold. This meticulous approach aims to ensure that the resulting models are trustworthy and capable of making accurate predictions. The quality of the final models was verified using the aforementioned test sets.

Statistical parameters, including sensitivity (SN), specificity (SP), and balanced accuracy (BA), were calculated to evaluate the predictive performance of the binary classifier (refer to Figure S1 in the Supporting Information). The balanced accuracy, occasionally referred to as the correct classification rate, is a metric that assesses the classification quality of the models. It is computed using the following formula:

BA = (SN + SP)/2 (1)

Balanced accuracy is further supported by a confusion matrix (see Figure S1), which illustrates the number of compounds accurately classified for each category and provides insights into misclassifications, such as the counts of false positives and false negatives.

In evaluating the performance of regression models, several key metrics were used to measure their accuracy and reliability. These metrics encompassed the root mean square error (RMSE), which quantifies the average deviation of the predicted values from the actual values; the mean absolute error (MAE), which measures the average absolute deviation between the predicted and actual values; the squared correlation coefficient (R²), a statistical measure that indicates the degree to which the model accurately fits the data; and the coefficient of determination (q²), which evaluates the predictive capacity of the model.[9] These metrics play a crucial role in assessing the accuracy and reliability of regression models, providing valuable insights that can guide further improvements or adjustments to the models.

OCHEM offers a valuable feature for estimating the applicability domain of developed models and the accuracy of forecasts, which is essential for ensuring their reliability and relevance.[39]

The platform’s manual provides in-depth information on the machine learning techniques utilized, the descriptors chosen, the statistical coefficients used, and the detailed validation procedures followed. This comprehensive documentation is beneficial for understanding and replicating the modeling process.[40]

 

Scroll to top

Synthesis of Oxazolo[4,5-d]pyrimidines

All reagents and solvents used in synthetic procedures were purchased from Aldrich and used as received. The reactions were followed by TLC (Silica gel, aluminum sheets 60 F₂₅₄, Merck). Melting points were recorded on a Fisher-Johns apparatus. ¹H and ¹³C NMR spectra were recorded on a Varian Mercury spectrometer (400 and 101 MHz, respectively) or Bruker Avance DRX 500 spectrometer (500 MHz and 126 MHz, respectively) in DMSO-d ₆, CDCl₃, or CF₃C(O)OD, taking its residual solvent signal as a standard. LC-MS analysis was performed on an Agilent 1200 Series system equipped with a diode array and a G6130A mass-spectrometer (atmospheric pressure electrospray ionization). Combustion elemental analysis was performed in the V.P. Kukhar Institute of Bioorganic Chemistry and Petrochemistry analytical laboratory; their results were found to be in good agreement (±0.4%) with the calculated values. Elemental analysis determined carbon and hydrogen contents using the Pregl gravimetric method, nitrogen using Duma’s gasometrical micromethod, and sulfur by the Scheininger titrimetric method.

 

Scroll to top

Synthesis of Oxazolo[4,5-d]pyrimidines 1–11; General Procedures

 

Scroll to top

Step 1: Synthesis of 7-Diazepane- or 7-Piperazine-Substituted Oxazolo[4,5-d]pyrimidines Va,b and VIa,b[41] [42]

A solution of 7-chlorooxazolo[4,5-d]pyrimidine IVa–c (0.01 mol) in anhyd dioxane (30 mL) was added dropwise to a solution of 1,4-diazepane or piperazine (0.03 mol) and Et₃N (0.01 mol) in dioxane (20 mL) over 0.5 h. The mixture was heated for 5 h at 100–110 °C and left for 12 h at 20–25 °C. The solvent was removed in vacuo; the residue was treated with water, and the solid precipitate was filtered off and dried, and the formed compounds Va,b or VIa,b were purified by crystallization (MeCN).

 

Scroll to top

Step 2: Synthesis of 7-Sulfonyldiazepane- or 7-Sulfonylpiperazine-Substituted Oxazolo[4,5-d]pyrimidines 1–11

To a solution of 7-diazepane- or 7-piperazine-substituted oxazolo[4,5-d]pyrimidines Va,b or VIa,b (0.001 mol) in anhydrous dioxane (20 mL) were added methane- or arenesulfonyl chloride (0.001 mol) and Et₃N (0.001 mol). The solution was refluxed for 2–3 h and left for 12 h at 20–25 °C; the solvent was removed, the residue was treated with water, the precipitate formed was filtered off, dried, and the obtained compound 1–11 was purified by crystallization (MeCN).

 

Scroll to top

2,5-Diphenyl-7-[4-(phenylsulfonyl)-1,4-diazepan-1-yl][1,3]oxazolo[4,5-d]pyrimidine (1)

Yield: 78%; white solid, mp 228–230 °С.

IR (KBr): 2936, 1619, 1548, 1480, 1445, 1369 (SO₂), 1320, 1149 (SO₂), 1118, 916, 766, 736, 694, 568 cm^–1.

¹H NMR (400 MHz, CDCl₃): δ = 8.46–8.45 (m, 2 H, ArH), 8.22 (d, 2 H, J _HH = 8.0 Hz, ArH), 7.73 (d, 2 H, J _HH = 8.0 Hz, ArH), 7.64–7.54 (m, 3 H, ArH), 7.47–7.41 (m, 4 H, ArH), 7.33 (d, 2 H, J _HH = 8.0 Hz, ArH), 4.21 (t, 2 H, J _HH = 4.0 Hz, CH₂ (diazepane)), 4.13 (t, 2 H, J _HH = 8.0 Hz, CH₂ (diazepane)), 3.64 (t, 2 H, J _HH = 4.0 Hz, CH₂ (diazepane)), 3.42 (t, 2 H, J _HH = 6.0 Hz, CH₂ (diazepane)), 2.21 (t, 2 H, J _HH = 4.0 Hz, CH₂ (diazepane)).

¹³C NMR (126 MHz, DMSO-d ₆): δ = 165.0, 161.8, 160.0, 147.7, 143.3, 138.4, 137.9, 133.2, 130.5, 130.0, 129.9, 129.8, 129.2, 128.8, 128.0, 126.8, 126.2, 50.6, 48.3, 46.2.

LC-MS: m/z (%) = 512.0 (100) [M + 1]⁺.

Anal. Calcd for C₂₈H₂₅N₅O₃S: C, 65.74; H, 4.93; N, 13.69; S, 6.27. Found: C, 65.42; H, 4.89; N, 13.33; S, 6.79.

 

Scroll to top

7-[4-(4-Methylphenylsulfonyl)-1,4-diazepan-1-yl]-2,5-diphenyl[1,3]oxazolo[4,5-d]pyrimidine (2)

Yield: 71%; white solid, mp 200–202 °С.

IR (KBr): 2934, 1619, 1548, 1480, 1448, 1393 (SO₂), 1319, 1148 (SO₂), 1117, 1074, 770, 697, 724, 542 cm^–1.

¹H NMR (400 MHz, CDCl₃): δ = 8.44 (br s, 2 H, ArH), 8.21 (d, 2 H, J _HH = 8.0 Hz, ArH), 7.63–7.55 (m, 5 H, ArH), 7.46–7.45 (m, 3 H, ArH), 7.05 (d, 2 H, J _HH = 4.0 Hz, ArH), 4.16 (br s, 2 H, CH₂ (diazepane)), 4.08 (t, 2 H, J _HH = 4.0 Hz, CH₂ (diazepane)), 3.62 (t, 2 H, J _HH = 4.0 Hz, CH₂ (diazepane)), 3.44 (t, 2 H, J _HH = 4.0 Hz, CH₂ (diazepane)), 2.27 (s, 3 H, Me), 2.16 (t, 2 H, J _HH = 4.0 Hz, CH₂ (diazepane)).

¹³C NMR (126 MHz, DMSO-d ₆): δ = 164.7, 161.7, 159.8, 147.7, 143.3, 138.4, 137.0, 133.2, 130.5, 129.9, 129.8, 128.8, 128.2, 128.1, 128.0, 126.8, 126.2, 50.1, 48.7, 46.9, 21.3.

LC-MS: m/z (%) = 526.2 (100) [M + 1]⁺.

Anal. Calcd for C₂₉H₂₇N₅O₃S: C, 66.27; H, 5.18; N, 13.32; S, 6.10. Found: C, 66.13; H, 5.28; N, 13.56; S, 6.45.

 

Scroll to top

2-(4-Methylphenyl)-7-[4-(methylsulfonyl)-1,4-diazepan-1-yl]-5-phenyl[1,3]oxazolo[4,5-d]pyrimidine (3)

Yield: 76%; white solid, mp 195–197 °С.

IR (KBr): 2923, 1617, 1550, 1486, 1372 (SO₂), 1325, 1153 (SO₂), 771, 701, 518 cm^–1.

¹H NMR (400 MHz, DMSO-d ₆): δ = 8.35–8.33 (m, 2 H, ArH), 8.08 (d, 2 H, J _HH = 8.0 Hz, ArH), 7.46–7.45 (m, 3 H, ArH), 7.41 (d, 2 H, J _HH = 8.0 Hz, ArH), 4.06–4.14 (m, 4 H, 2 CH₂ (diazepane)), 3.59 (t, 2 H, J _HH = 4.0 Hz, CH₂ (diazepane)), 3.32 (br s, 2 H, CH₂ (diazepane)), 2.83 (s, 3 H, MeSO₂), 2.39 (s, 3 H, Me), 2.01 (t, 2 H, J _HH = 4.0 Hz, CH₂ (diazepane)).

¹³C NMR (126 MHz, DMSO-d ₆): δ = 143.7, 138.5, 137.6, 137.2, 130.5, 130.5, 130.4, 128.9, 128.2, 128.2, 123.5, 122.1, 47.1, 37.5, 37.1, 21.8.

LC-MS: m/z (%) = 464.0 (100) [M + 1]⁺.

Anal. Calcd for C₂₄H₂₅N₅O₃S: C, 62.19; H, 5.44; N, 15.11; S, 6.92. Found: C, 62.01; H, 5.41; N, 15.34; S, 6.77.

 

Scroll to top

2-(4-Methylphenyl)-5-phenyl-7-[4-(phenylsulfonyl)-1,4-diazepan-1-yl][1,3]oxazolo[4,5-d]pyrimidine (4)

Yield: 77%; white solid, mp 229–231 °С.

IR (KBr): 2941, 1618, 1550, 1488, 1371 (SO₂), 1153 (SO₂), 1117, 733, 692, 572 cm^–1.

¹H NMR (400 MHz, DMSO-d ₆): δ = 8.31–8.30 (m, 2 H, ArH). 8.05 (d, 2 H, J _HH = 8.0 Hz, ArH), 7.65 (d, 2 H, J _HH = 8.0 Hz, ArH), 7.45–7.33 (m, 8 H, ArH), 4.09 (t, 2 H, J _HH = 6.0 Hz, CH₂ (diazepane)), 3.97 (t, 2 H, J _HH = 6.0 Hz, CH₂ (diazepane)), 3.60–3.59 (m, 2 H, CH₂ (diazepane)), 3.39 (br s, 2 H, CH₂ (diazepane)), 2.38 (s, 3 H, Me), 1.96 (br s, 2 H, CH₂ (diazepane)).

¹³C NMR (126 MHz, DMSO-d ₆): δ = 165.0, 162.0, 161.0, 160.0, 159.8, 143.7, 139.7, 138.4, 134.1, 133.0, 130.4, 129.5, 128.8, 128.2, 128.1, 126.9, 123.5, 49.3, 47.6, 47.4, 21.7.

LC-MS: m/z (%) = 526.2 (100) [M + 1]⁺.

Anal. Calcd for C₂₉H₂₇N₅O₃S: C, 66.27; H, 5.18; N, 13.32; S, 6.10. Found: C, 66.13; H, 5.42; N, 13.56; S, 6.32.

 

Scroll to top

2-(4-Methylphenyl)-7-[4-(4-methylphenylsulfonyl)-1,4-diazepan-1-yl]-5-phenyl[1,3]oxazolo[4,5-d]pyrimidine (5)

Yield: 77%; white solid, mp 192–194 °С.

IR (KBr): 3060, 2946, 2914, 1609, 1552, 1489, 1372 (SO₂), 1335, 1157 (SO₂), 1104, 714, 697, 549 cm^–1.

¹H NMR (400 MHz, CDCl₃): δ = 8.46–8.45 (m, 2 H, ArH), 8.11 (d, 2 H, J _HH = 8.0 Hz, ArH), 7.57 (d, 2 H, J _HH = 8.0 Hz, ArH), 7.47–7.46 (m, 3 H, ArH), 7.37 (d, 2 H, J _HH = 8.0 Hz, ArH), 7.06 (d, 2 H, J _HH = 8.0 Hz, ArH), 4.19 (t, 2 H, J _HH = 6.0 Hz, CH₂ (diazepane)), 4.11 (t, 2 H, J _HH = 6.0 Hz, CH₂ (diazepane)), 3.63 (t, 2 H, J _HH = 6.0 Hz, CH₂ (diazepane)), 3.44 (t, 2 H, J _HH = 6.0 Hz, CH₂ (diazepane)), 2.48 (s, 3 H, Me), 2.28 (s, 3 H, Me), 2.21–2.16 (m, 2 H, CH₂ (diazepane)).

¹³C NMR (101 MHz, CDCl₃): δ = 160.6, 159.9, 147.2, 143.5, 137.9, 136.4, 130.1, 129.9, 129.8, 129.5, 128.3, 128.2, 128.1, 127.9, 127.8, 126.7, 123.4, 21.8, 21.4.

LC-H: m/z (%) = 540.0 (100) [M + 1]⁺.

Anal. Calcd for C₃₀H₂₉N₅O₃S: C, 66.77; H, 5.42; N, 12.98; S, 5.94. Found: C, 66.54; H, 5.62; N, 12.65; S, 5.67.

 

Scroll to top

7-[4-(Methylsulfonyl)piperazin-1-yl]-2,5-diphenyl[1,3]oxazolo[4,5-d]pyrimidine (6)

Yield: 76%; white solid, mp 267–269 °С.

IR (KBr): 2927, 1619, 1548, 1477, 1447, 1374 (SO₂), 1326, 1159 (SO₂), 961, 773, 699, 661, 516 cm^–1.

¹H NMR (400 MHz, CDCl₃): δ = 8.48–8.46 (m, 2 H, ArH), 8.24 (d, 2 H, J _HH = 8.0 Hz, ArH), 7.64–7.55 (m, 3 H, ArH), 7.48–7.46 (m, 3 H, ArH), 4.24 (t, 4 H, J _HH = 4.0 Hz, 2 CH₂(piperazine)), 3.49 (t, 4 H, J _HH = 4.0 Hz, 2 CH₂(piperazine)), 2.87 (s, 3 H, Me).

¹³C NMR (101 MHz, CDCl₃): δ = 160.8, 137.7, 132.8, 130.3, 129.7, 129.3, 129.2, 128.4, 128.3, 128.2, 128.1, 125.9, 45.6, 44.8, 34.8.

LC-MS: m/z (%) = 436.0 (100) [M + 1]⁺.

Anal. Calcd for C₂₂H₂₁N₅O₃S: C, 60.68; H, 4.86; N, 16.08; S, 7.36. Found: C, 60.56; H, 4.98; N, 16.38; S, 7.59.

 

Scroll to top

2,5-Diphenyl-7-[4-(phenylsulfonyl)piperazin-1-yl][1,3]oxazolo[4,5-d]pyrimidine (7)

Yield: 71%; white solid, mp 260–262 °С.

IR (KBr): 2861, 1621, 1549, 1477, 1448, 1373 (SO₂), 1349, 1314, 1170 (SO₂), 938, 925, 742, 697, 574 cm^–1.

¹H NMR (500 MHz, DMSO-d ₆): δ = 8.31 (d, 2 H, ArH, J _HH = 6.0 Hz), 8.23 (d, 2 H, J _HH = 4.5 Hz, ArH), 7.80–7.79 (m, 2 H, ArH), 7.69–7.62 (m, 6 H, ArH), 7.45 (br s, 3 H, ArH), 4.13 (br s, 4 H, 2 CH₂ (piperazine)), 3.20 (br s, 4 H, 2 CH₂ (piperazine)).

¹³C NMR (126 MHz, CF₃C(O)OD): δ = 169.3, 154.7, 150.8, 147.6, 135.4, 134.2, 133.9, 132.7, 129.1, 129.0, 128.9, 127.9, 127.1, 126.5, 126.5, 121.4, 46.7, 44.8, 44.7, 43.7.

LC-MS: m/z (%) = 498.0 (100) [M + 1]⁺.

Anal. Calcd for C₂₇H₂₃N₅O₃S: C, 65.18; H, 4.66; N, 14.08; S, 6.44. Found: C, 65.31; H, 4.55; N, 14.44; S, 6.79.

 

Scroll to top

7-[4-(4-Methylphenylsulfonyl)piperazin-1-yl]-2,5-diphenyl[1,3]oxazolo[4,5-d]pyrimidine (8)

Yield: 69%; white solid, mp 280–282 °С.

IR (KBr): 2857, 1616, 1549, 1476, 1446, 1375 (SO₂), 1347, 1166 (SO₂), 939, 769, 727, 546 cm^–1.

¹H NMR (400 MHz, CDCl₃): δ = 8.45–8.43 (m, 2 H, J _HH = 8.0 Hz, ArH), 8.24 (d, 2 H, J _HH = 8.0 Hz, ArH), 7.68 (d, 2 H, J _HH = 8.0 Hz, ArH), 7.62 (t, 1 H, J _HH = 8.0 Hz, ArH), 7.56 (t, 2 H, J _HH = 8.0 Hz, ArH), 7.45 (d, 3 H, J _HH = 8.0 Hz, ArH), 7.33 (d, 2 H, J _HH = 8.0 Hz, ArH), 4.23 (t, 4 H, J _HH = 4.0 Hz, 2 CH₂ (piperazine)), 3.23 (t, 4 H, J _HH = 4.0 Hz, 2 CH₂ (piperazine)), 2.40 (s, 3 H, Me),

¹³C NMR (126 MHz, DMSO-d ₆): δ = 133.4, 130.7, 130.6, 130.5, 130.4, 129.8, 129.7, 128.9, 128.8, 128.6, 128.4, 128.3, 128.3, 128.1, 128.0, 126.1, 126.0, 58.6, 53.3, 50.7, 46.2, 21.5.

LC-MS: m/z (%) = 512.0 (100) [M + 1]⁺.

Anal. Calcd for C₂₈H₂₅N₅O₃S: C, 65.74; H, 4.93; N, 13.69; S, 6.27. Found: C, 65.62; H, 4.89; N, 13.33; S, 6.49.

 

Scroll to top

2-(4-Methylphenyl)-7-[4-(methylsulfonyl)piperazin-1-yl]-5-phenyl[1,3]oxazolo[4,5-d]pyrimidine (9)

Yield: 72%; white solid, mp 288–290 °С.

IR (KBr): 2990, 2844, 1608, 1550, 1443, 1372, 1323, 1161, 947, 776, 692 cm^–1.

¹H NMR (400 MHz, DMSO-d ₆): δ = 8.33–8.32 (m, 2 H, ArH), 8.11 (d, 2 H, J _HH = 8.0 Hz, ArH), 7.45 (s, 3 H, ArH), 7.39, (d, 2 H, J _HH = 8.0 Hz, ArH), 4.12 (br s, 4 H, 2 CH₂ (piperazine)), 3.34 (br s, 4 H, 2 CH₂ (piperazine)), 2.89 (s, 3 H, Me), 2.38 (s, 3 H, Me).

LC-MS: m/z (%) = 452.0 (100) [M + 1]⁺.

Anal. Calcd for C₂₃H₂₃N₅O₃S: C, 61.45; H, 5.16; N, 15.58; S, 7.13. Found: C, 61.35; H, 5.25; N, 15.78; S, 7.46.

 

Scroll to top

2-(4-Methylphenyl)-5-phenyl-7-[4-(phenylsulfonyl)piperazin-1-yl][1,3]oxazolo[4,5-d]pyrimidine (10)

Yield: 76%; white solid, mp 247–249 °С.

IR (KBr): 3610, 2862, 1617, 1478, 1447, 1376 (SO₂), 1166 (SO₂), 1120, 949, 741, 693, 576 cm^–1.

¹H NMR (400 MHz, CDCl₃): δ = 8.45–8.42 (m, 2 H, ArH), 8.11 (d, 2 H, J _HH = 8.0 Hz, ArH), 7.81 (d, 2 H, J _HH = 8.0 Hz, ArH), 7.62–7.53 (m, 3 H, ArH), 7.45–7.44 (m, 3 H, ArH), 7.35 (d, 2 H, J _HH = 8.0 Hz, ArH), 4.22 (t, 4 H, J _HH = 6.0 Hz, 2 CH₂ (piperazine)), 3.26 (t, 4 H, J _HH = 6.0 Hz, 2 CH₂ (piperazine)), 2.47 (s, 3 H, Me).

¹³C NMR (101 MHz, CF₃C(O)OD): δ = 170.1, 154.9, 151.3, 149.0, 147.9, 134.7, 134.5, 133.0, 130.5, 129.5, 128.6, 128.4, 127.6, 127.1, 126.7, 126.0, 118.9, 45.3, 44.6, 44.1, 43.4, 20.8.

LC-MS: m/z (%) = 512.0 (100) [M + 1]⁺.

Anal. Calcd for C₂₈H₂₅N₅O₃S: C, 65.74; H, 4.93; N, 13.69; S, 6.27. Found: C, 65.56; H, 4.78; N, 13.99; S, 6.46.

 

Scroll to top

2-(4-Methylphenyl)-7-[4-(4-methylphenylsulfonyl)piperazin-1-yl]-5-phenyl[1,3]oxazolo[4,5-d]pyrimidine (11)

Yield: 71%; white solid, mp 261–263 °С.

IR (KBr): 2860, 1613, 1580, 1551, 1482, 1446, 1397, 1373 (SO₂), 1347, 1327, 1313, 1264, 1214, 1163 (SO₂), 1117, 1096, 1057, 1018, 946, 848, 821, 772, 725, 702, 651, 612, 577, 548 cm^–1.

¹H NMR (400 MHz, DMSO-d ₆): δ = 8.32–8.31 (m, 2 H, ArH), 8.10 (d, 2 H, J _HH = 7.4 Hz, ArH), 7.69 (d, 2 H, J _HH = 7.6 Hz, ArH), 7.45–7.41 (m, 7 H, ArH), 4.12 (br s, 4 H, 2 CH₂ (piperazine)), 3.24 (br s, 4 H, 2 CH₂ (piperazine)), 2.42 (s, 3 H, Me), 2.36 (s, 3 H, Me).

LC-MS: m/z (%) = 526.0 (100) [M + 1]⁺.

Anal. Calcd for C₂₉H₂₇N₅O₃S: C, 66.27; H, 5.18; N, 13.32; S, 6.10. Found: C, 66.11; H, 5.13; N, 13.43; S, 6.31.

 

Scroll to top

In vitro Anticancer Screening of the Tested Compounds

 

Scroll to top

One-Dose Assay

Synthesized compounds were submitted to National Cancer Institute NCI, Bethesda, Maryland, U.S.A. under the Developmental Therapeutic Program DTP. A primary in vitro one-dose anticancer assay was performed using the total cell line panel by the protocol of the Drug Evaluation Branch, NCI, Bethesda.[43] [44] [45] The compounds were added at a single concentration (10^–5 M), and the culture was incubated for 48 h. Endpoint determinations were made with a protein-binding dye, sulforhodamine B (SRB). Results for each compound were reported as the percent of growth of the treated cells compared to the untreated control cells.

 

Scroll to top

COMPARE Correlations

The graph of mean values for each of the compounds was subsequently used to run the COMPARE algorithm from the Developmental Therapeutics Program, NCI, and calculate the correlation coefficient with respect to compounds from the standard agent database with a known mechanism of action.[46] A pairwise correlation coefficient ≥ 0.70 was used as the cut-off for assessing whether two agents were likely to share a similar mechanism of action. Briefly, vectors of GI₅₀, TGI, and LC₅₀ concentrations for tested compounds were correlated with the set of average GI₅₀, TGI, and LC₅₀ vectors for all public NCI-60 vectors for the full public standard agents database.[47]

 

Scroll to top

ADMET Analysis

Available online websites, such as ADMETlab 3.0, an integrated online platform for Windows,[48] SwissADME,[49] and Deep-PK,[50] were applied to explore ADMET properties of the studied molecules. The pharmacokinetic, pharmacodynamic, and toxic targets of the test compounds were predicted. The conversion of molecular structures into SMILES strings, which is required for the operation of ADMET platforms, was done using the Marvin JS widget.[51]

 

Scroll to top

Molecular Docking Study

AutoDock Tools (ADT) 1.5.7 was used to prepare the protein and ligand.[52] In the 3D protein molecule, hydrogen atoms were added only for polar atoms, and all atoms were again renumbered. The partial charges were calculated using the Gasteiger method. The ligand structure and pre-conformation (.mol format) were constructed using the ChemAxon Marvin Sketch v.5.5.01 software.[53] Their structures were optimized and minimized energy (by Avogadro v.1.2.0).[54] The AutoDock Vina v.1.1.2 was employed for molecular docking.[55] The docking center was: x = –4.178, y = 4.737, and z = 14.218, and a grid box of size 35 × 32 × 30 and a grid step of 1.0 Å was used. Accelrys Studio Visualizer (v 19.1.0) was used for interaction analysis and result visualization.[56]

 

Scroll to top

 

Scroll to top

Conflict of Interest

The authors declare no conflict of interest.

Acknowledgment

We thank the National Cancer Institute, Bethesda, MD, USA, for the in vitro evaluation of anticancer activity within the Developmental Therapeutic Program and Enamine Ltd for the material and technical support.

• References

• 1 Sochacka-Ćwikła A, Mączyński M. Molecules 2025; 30: 666
• 2 De Coen LM, Roman BI, Movsisyan M, Heugebaert TS. A, Stevens CV. Eur. J. Org. Chem. 2018; 2018: 2148
• 3 Velihina Y, Gesese R, Zhirnov V, Kobzar O, Bui B, Pilyo S, Vovk A, Shen HY, Brovarets V. RSC Med. Chem. 2023; 14: 692
• 4 Karatas E, Foto E, Ertan-Bolelli T, Yalcin-Ozkat G, Yilmaz S, Ataei S, Zilifdar F, Yildiz I. Bioorg. Chem. 2021; 112: 104913
• 5 Zhao C, Liu Y, Cui Z. Future Med. Chem. 2024; 16: 1267
• 6 Souza JM. T, Silva SA. N. M, Rocha RB. D, Machado FD. S, Marinho Filho JD. B, Araújo AJ. Cell Biochem. Funct. 2024; 42: e70001
• 7 Poliudov AO, Havryshko DY, Sorokin MD, Yatsymyrskyi AV, Virych PA, Shishkina SV, Fokin AA, Omelian TV, Kysil AI, Milokhov DS, Dobrydnev AV. J. Mol. Struct. 2025; 1325: 140952
• 8 Sushko I, Novotarskyi S, Körner R, Pandey AK, Rupp M, Teetz W, Brandmaier S, Abdelaziz A, Prokopenko VV, Tanchuk VY, Todeschini R, Varnek A, Marcou G, Ertl P, Potemkin V, Grishina M, Gasteiger J, Schwab C, Baskin II, Palyulin VA, Radchenko EV, Welsh WJ, Kholodovych V, Chekmarev D, Cherkasov A, Aires-de-Sousa J, Zhang Q.-Y, Bender A, Nigsch F, Patiny L, Williams A, Tkachenko V, Tetko IV. J. Comput.-Aided Mol. Des. 2011; 25: 533
• 9 Jonhos NN, Leone FG. Statistic and Experimental Design in Engineering and the Physical Sciences. John Wiley & Sons; New York: 1977
• 10 Li ZH, Liu XQ, Geng PF, Ma JL, Zhao TQ, Wei HM, Yu B, Liu HM. MedChemComm 2017; 8: 1655
• 11 Gryniukova A, Borysko P, Myziuk I, Alieksieieva D, Hodyna D, Semenyuta I, Kovalishyn V, Metelytsia L, Rogalsky S, Tcherniuk S. Mol. Diversity 2024; 28: 3817
• 12 Sviripa VM, Gakh AA, Brovarets VS, Gutov AV, Drach BS. Synthesis 2006; 3462
• 13 Acton EM, Narayanan VL, Risbood PA, Shoemaker RH, Vistica DT, Boyd MR. J. Med. Chem. 1994; 37: 2185
• 14 Holbeck SL, Collins JM, Doroshow JH. Mol. Cancer Ther. 2010; 9: 1451
• 15 Li P, Hua L, Ma Z, Hu W, Liu Y, Zhu J. J. Chem. Inf. Model. 2024; 64: 8705
• 16a FAFDrugs4: https://fafdrugs4.rpbs.univ-paris-diderot.fr/index.html (accessed Sept 16, 2025).
• 16b Summerfield SG, Read K, Begley DJ, Obradovic T, Hidalgo IJ, Coggan S, Lewis VA, Porter RA, Jeffrey P. J. Pharm. Exp. Ther. 2007; 322: 205
• 17 ADMETlab 2.0: an integrated online platform for accurate and comprehensive predictions of ADMET properties.
• 18 Xiong G, Wu Z, Yi J, Fu L, Yang Z, Hsieh C, Yin M, Zeng X, Wu C, Lu A, Chen X, Hou T, Cao D. Nucleic Acids Res. 2021; 49: W5
• 19 Zhao M, Ma J, Li M, Zhang Y, Jiang B, Zhao X, Huai C, Shen L, Zhang N, He L, Qin S. Int. J. Mol. Sci. 2021; 22: 12808
• 20 Villa M, Wu J, Hansen S, Pahnke J. Cells 2024; 13: 740
• 21 Nanayakkara AK, Follit CA, Chen G, Williams NS, Vogel PD, Wise JG. Sci. Rep. 2018; 8: 967
• 22 Heymans M, Sevin E, Gosselet F, Lundquist S, Culot M. Eur. J. Pharm. Biopharm. 2018; 127: 453
• 23 Ivanenkov YA, Zagribelnyy BA, Aladinskiy VA. J. Med. Chem. 2019; 62: 10026
• 24 Grave N, Scheffel TB, Cruz FF, Rockenbach L, Goettert MI, Laufer S, Morrone FB. Front. Pharmacol. 2022; 13: 975197
• 25 Haapasalo J, Nordfors K, Haapasalo H, Parkkila S. Cancers 2020; 12: 1723
• 26 Berman HM, Westbrook J, Feng Z, Gilliland G, Bhat TN, Weissig H, Shindyalov IN, Bourne PE. Nucleic Acids Res. 2000; 28: 235
• 27 Mendez D, Gaulton A, Bento AP, Chambers J, De Veij M, Félix E, Magariños MP, Mosquera JF, Mutowo P, Nowotka M, Gordillo-Marañón M, Hunter F, Junco L, Mugumbate G, Rodriguez-Lopez M, Atkinson F, Bosc N, Radoux CJ, Segura-Cabrera A, Hersey A, Leach AR. Nucleic Acids Res. 2019; 47: D930
• 28 Instant JChem; http://www.chemaxon.com/products/instant-jchem/ (accessed on Feb 10, 2025).
• 29 Kennard RW, Stone LA. Technometrics 1969; 11: 137
• 30 OCHEM; https://ochem.eu/ (accessed March 10, 2025).
• 31 Karpov P, Godin G, Tetko IV. J. Cheminf. 2020; 12: 17
• 32 Tetko IV, Karpov P, Bruno E, Kimber TB, Godin G. Augmentation Is What You Need! . In Artificial Neural Networks and Machine Learning – ICANN 2019: Workshop and Special Sessions . Tetko I, Kůrková V, Karpov P, Theis F. Lecture Notes in Computer Science 11731; Springer; Cham: 2019. DOI org/10.1007/978-3-030-30493-5_79
• 33 Tetko IV. Methods Mol. Biol. 2008; 458: 185
• 34 Breiman L. Mach. Learn. 2001; 45: 5
• 35 Hall LH, Kier LB. J. Chem. Inf. Comput. Sci. 1995; 35: 1039
• 36 Tetko IV, Tanchuk VY. J. Chem. Inf. Comput Sci. 2002; 42: 1136
• 37 Willighagen EL, Mayfield JW, Alvarsson J, Berg A, Carlsson L, Jeliazkova N, Kuhn S, Pluskal T, Rojas-Chertó M, Spjuth O, Torrance G, Evelo CT, Guha R, Steinbeck C. J. Cheminf. 2017; 9: 33
• 38 Tetko IV, Sushko I, Pandey AK, Zhu H, Tropsha A, Papa E, Oberg T, Todeschini R, Fourches D, Varnek A. J. Chem. Inf. Model. 2008; 48: 1733
• 39 Sushko I, Novotarskyi S, Körner R, Pandey AK, Kovalishyn VV, Prokopenko VV, Tetko IV. J. Chemom. 2010; 24: 202
• 40 OCHEM user’s manual; http://docs.ochem.eu/display/MAN (accessed Feb 10, 2025).
• 41 Velihina YS, Kachaeva MV, Pilyo SG, Mitiukhin OP, Zhirnov VV, Brovarets VS. J. Chem. Res. 2018; 3: 81
• 42 Velihina YS, Kachaeva MV, Pilyo SG, Zhirnov VV, Brovarets VS. Pharma Chem. 2018; 10: 1
• 43 Grever MR, Schepartz SA, Chabner BA. Semin. Oncol. 1992; 19: 622
• 44 Boyd MR, Paull KD. Drug. Rev. Res. 1995; 34: 91
• 45 Monks A, Scudiero D, Skehan P, Shoemaker R, Paull K, Vistica D, Hose C, Jangley J, Cronisie P, Viagro-Wolff A, Gray-Goodrich M, Campell H, Boyd M. J. Natl. Cancer Inst. 1991; 83: 757
• 46 COMPARE Analysis; https://dtp.cancer.gov/databases_tools/compare.htm, (accessed Mar 20, 2025).
• 47 Shoemaker RH. Nat. Rev. Cancer 2006; 6: 813
• 48 ADMET Evaluation - ADMETlab 3.0 - CBDD-Group; https://admetlab3.scbdd.com/server/evaluation (accessed Mar 20, 2025).
• 49 SwissADME; http://www.swissadme.ch/index.php (accessed Mar 20, 2025).
• 50 Deep-PK; https://biosig.lab.uq.edu.au/deeppk/prediction (accessed Mar 20, 2025).
• 51 Marvin JS widget; https://docs.chemaxon.com/display/lts-europium/introduction-to-marvinview.md (accessed Mar 20, 2025).
• 52 Morris GM, Huey R, Lindstrom W, Sanner MF, Belew RK, Goodsell DS, Olson AJ. J. Comput. Chem. 2009; 30: 2785
• 53 ChemAxon, Marvin Sketch 5.5.0.1; https://www.chemaxon.com (accessed Marсh 20, 2025).
• 54 Hanwell MD, Curtis DE, Lonie DC, Vandermeersch T, Zurek E, Hutchison GR. J. Cheminf. 2012; 4: 17
• 55 Trott O, Olson AJ. J. Comput. Chem. 2010; 31: 455
• 56 Dassault Systèmes, Discovery Studio Visualizer; https://discover.3ds.com/ (accessed Mar 20, 2025).

Publikationsverlauf

Eingereicht: 07. Juli 2025

Angenommen nach Revision: 19. August 2025

Accepted Manuscript online:
21. August 2025

Artikel online veröffentlicht:
13. Oktober 2025