4.2.2. Synthesis of imidazolium-based ILs
(compound 2) 1-Dodecyl-3-methylimidazolium chloride (C12C1IM-Cl) was synthesized according to Scheme 2. The stirred mixture of 1-methylimidazole (5 g, 0.06 mol) and 1-chlorododecane (15 g, 0.07 mol) was heated at 140°C for 24 h. After cooling to room temperature, the solid product was washed with hexane-ethyl acetate mixture (3:1 (v/v), 3 x 50 ml). Residual solvents were removed under vacuum (10 mbar) at 60°C.
Yield: 65% (11 g), white solid, mp 46–47°C.
1H NMR (400 МHz, CDCl3): δ = 0.84 (t, J = 6.7 Hz, 3H, CH3), 1.14–1.36 (m, 18H, СH3(CH2)9), 1.87 (t, J = 7.3 Hz, 2H, NCH2CH2), 4.10 (s, 3H, NCH3), 4.23–4.32 (m, 2H, NCH2), 7.32 (t, J = 1.8 Hz, 1H, C4-H), 7.49 (t, J = 1.8 Hz, 1H, C5-H), 10.58 (d, J = 1.6 Hz, 1H, C2-H).
Long-chain ester-functionalized imidazolium ILs were synthesized as follows (Scheme 2). The mixture of 1-methylimidazole and alkyl chloroacetate (10% molar excess) was stirred at 120°C for 2 h. After cooling, the solid products were obtained. They were purified by recrystallization from ethyl acetate (IMC1CH2COOC12-Cl) or washed with hexane-ethyl acetate mixture (3:1 (v/v), IMC1CH2COOC10-Cl).
(compound 4) 1-Decyloxycarbonylmethyl-3-methylimidazolium chloride (IMC1CH2COOC10-Cl)
Yield: 72%, white solid, mp 74–76 oC
1H NMR (400 MHz, CDCl3): δ = 0.83 (t, J = 6.6 Hz, 3H, CH3), 1.12–1.32 (m, 14H, (CH2)7), 1.59 (q, J = 7.0 Hz, 2H, COOCH2CH2), 4.04 (s, 3H, NCH3), 4.11 (t, J = 6.9 Hz, 2H, COOCH2), 5.43 (s, 2H, NCH2CO), 7.54 (t, J = 1.8 Hz, 1H, C4-H), 7.60 (t, J = 1.8 Hz, 1H, C5-H), 10.38 (d, J = 1.6 Hz, 1H, C2-H).
(compound 9) 1-Dodecyloxycarbonylmethyl-3-methylimidazolium chloride (IMC1CH2COOC12-Cl)
Yield: 80%, white solid, mp 56–58 oC
1H NMR (400 MHz, CDCl3): δ = 0.86 (t, J = 6.9 Hz, 3H, CH3), 1.12–1.31 (m, 18H, (CH2)9), 1.56 (p, J = 6.9 Hz, 2H, COOCH2CH2), 4.01 (s, 3H, NCH3), 4.08 (t, J = 6.9 Hz, 2H, COOCH2), 5.42 (s, 2H, NCH2CO), 7.57 (t, J = 1.8 Hz, 1H, C4-H), 7.62 (t, J = 1.8 Hz, 1H, C5-H), 10.29 (d, J = 1.8 Hz, 1H, C2-H).
4.2.3. Synthesis of imidazole-based lysosomotropic detergents
(compound 7) 1-Dodecylimidazole (IMC12) was synthesized according to Scheme 3. Sodium hydride (3.5 g, 60% suspension in mineral oil) was washed with dry hexane on the filter and then suspended in 50 ml of acetonitrile. The reaction flask was placed in an ice bath. To the stirred mixture, imidazole (5 g, 0.07 mol) was added in small amounts. The reaction was allowed to continue for 2 h. Then 1-chlorododecane (14 g, 0.07 mol) was added and the mixture was refluxed for 12 h. After cooling to room temperature, it was poured into water (150 ml). The former organic layer was separated and dissolved in 100 ml of 1 M hydrochloric acid. The solution was washed with hexane (2 x 100 ml) to remove unreacted 1-chlorododecane. The 1-dodecylimidazole hydrochloride was converted to free base by the addition of 50 ml of 2 M NaOH. The oily product was extracted with methylene chloride (2 x 50 ml) and dried over sodium sulfate. The methylene chloride was distilled off at atmospheric pressure. Residual solvent was removed under vacuum 10 mbar at 50°C.
Yield: 60% (10 g), yellow liquid.
1H NMR (400 МHz,CDCl3): δ = 0.87 (t, 3H, CH3), 1.17–1.33 (m, 18H, СH3(CH2)9), 1.75 (t, J = 7.0 Hz, 2H, NCH2CH2), 3.9 (t, J = 7.2 Hz, 2H, NCH2), 6.88 (t, J = 1.3 Hz, 1H, C4-H), 7.03 (d, J = 1.2 Hz, 1H, C5-H), 7.44 (d, J = 1.2 Hz, 1H, C2-H).
1-alkyloxycarbonylmethylimidazoles were synthesized using similar approach (Scheme 3). The mixture of sodium imidazole (0.07 mol) and corresponding alkyl chloroacetate (0.07 mol) in acetonitrile (50 ml) was refluxed for 6 h. After cooling to room temperature, it was poured into water (150 ml). The formed precipitate was filtered and dried in vacuum 10 mbar at 50°C. Finally, it was purified by recrystallization from hexane.
(compound 8) 1-Dodecyloxycarbonylmethylimidazole (IMCH2COOC12)
Yield: 75% (15.4 g), white solid, mp 94–95°C
1H NMR (400 МHz,CDCl3): δ = 0.87 (t, J = 6.7 Hz, 3H, CH3), 1.18–1.35 (m, 18H, СH3(CH2)9), 1.63 (t, J = 6.9 Hz, 2H, NCH2CH2), 4.16 (t, J = 6.7 Hz, 2H, COOCH2), 4.69 (s, 2H, NCH2), 6.95 (d, J = 1.3 Hz, 1H, C4-H), 7.10 (s, 1H, C5-H), 7.51 (s, 1H, C2-H).
(compound 3) 1-Decyloxycarbonylmethylimidazole (IMCH2COOC10)
Yield: 81% (15 g), white solid, mp 85°C
1H NMR (400 МHz,CDCl3): δ = 0.86 (t, J = 6.7 Hz, 3H, CH3), 1.18–1.36 (m, 14H, СH3(CH2)7), 1.62 (q, J = 6.9 Hz, 2H, NCH2CH2), 4.16 (t, J = 6.7 Hz, 2H, COOCH2), 4.69 (s, 2H, NCH2), 6.95 (d, J = 1.3 Hz, 1H, C4-H), 7.09 (s, 1H, C5-H), 7.50 (s, 1H, C2-H).
QSAR modeling
The data for our analysis were obtained from several publications. For QSAR modelling, the data were uploaded to the On-line Chemical Database and Modelling Environment (OCHEM) database [57].
Two data sets were used to build the QSAR models. The first dataset included 441 compounds and their bioactivities against various neuroblastoma cell lines, with the cytotoxicity of the compounds, expressed as IC50 ranging from 0.0000278 to 252 µM. The second data set (8214 compounds) consisted of various chemical series with IC50 values of the molecules ranging from 0.0000284 to 30000 µM against the leukaemia cell line K-562. The initial data obtained as IC50 were converted to log (1/IC50) values and used as the dependent variable for QSAR modeling.
Five Machine Learning Techniques (MLTs) such as Transformer Convolutional Neural Network (Trans-CNN) [58], Message Passing Neural Networks (MPNN) [59], Transformer Convolutional Neural Fingerprint (Trans-CNF) [60], Random Forest Regression (RFR) [61], and Deep Learning Consensus Architecture (DLCA) [62] were used to build models. The MLTs are described in detail in the Supplemental materials.
The quality of the models was evaluated using fivefold cross-validation (CV) with variable selection in each step and an external test set [63]. The OCHEM also provides an estimation of the domain of applicability [64] of the developed QSAR models and the accuracy of predictions.
We used two criteria to assess the goodness of fit: the squared correlation coefficient R2 and the coefficient of determination q2. In addition, the root mean square error (RMSE) and the mean absolute error (MAE) were calculated to estimate the prediction errors.