Ligand-based pharmacophore modeling and Bayesian approaches to identify c-Src inhibitors

Figures & data

Figure 1. Overall scheme of the work.

Figure 2. Chemically diverse 33 compounds used as training set in 3D-QSAR Pharmacophore generation (Discovery Studio). IC₅₀ values are indicated in parentheses for each compound.

Table 1. Information of statistical significance values are presented in cost values measured in bits for the top 10 hypotheses as a result of automated 3D-QSAR pharmacophore generation.

Hypo no.	Total cost	Cost difference*	RMS†	Correlation	Features†	Max. fit
Hypo1	135.76	90.52	0.64	0.97	HBA, HBD, Hy-Ar, RA	10.46
Hypo2	137.58	88.7	0.75	0.95	HBA, HBD, Hy-Ar, RA	9.98
Hypo3	138.26	88.02	0.69	0.96	HBA, HBD, Hy-Ar, RA	10.16
Hypo4	140.61	85.67	0.87	0.95	HBA, HBD, Hy-Ar, RA	10
Hypo5	142.09	84.19	0.95	0.93	HBA, HBD, Hy-Ar, RA	8.86
Hypo6	142.09	84.19	0.95	0.93	HBA, HBD, Hy-Ar, RA	8.58
Hypo7	142.66	83.62	0.97	0.93	HBA, HBD, Hy-Ar, RA	8.96
Hypo8	142.79	83.49	0.98	0.93	HBA, HBD, Hy-Ar, RA	8.53
Hypo9	143.77	82.51	1.01	0.92	HBA, HBD, Hy-Ar, RA	8.2
Hypo10	144.24	82.04	1.02	0.92	HBA, HBD, Hy-Ar, RA	8.78

*Cost difference between the null and the total cost. The null cost, the fixed cost and the configuration cost are 226.28, 126.83 and 14.71, respectively.

†RMS, root mean square deviation; HBA, hydrogen bond acceptor; HBD, hydrogen bond donor; Hy-Ar, hydrophobic aromatic; RA, ring aromatic.

Figure 3. Hypo1 pharmacophore model with its geometric constrains: hydrogen bond acceptor (HBA, green), hydrogen bond donor (HBD, magenta), hydrophobic aromatic (Hy-Ar, blue) and ring aromatic (RA, brown).

Table 2. Distance between every two features of Hypo1.

Hypo1	HBA	HBD	Hy-Ar	RA
HBA	0	4.663	2.989	5.231
HBD	4.663	0	3.179	3.105
Hy-Ar	2.989	3.179	0	4.663

HBA, hydrogen bond acceptor; HBD, hydrogen bond donor; Hy-Ar, hydrophobic aromatic; RA, ring aromatic.

Table 3. Actual and estimated activity values of the training set molecules based on the pharmacophore model Hypo1.

Compound no.	Exp. IC50 nM	Pred. IC50 nM	Error*	Fit value†	Exp. scale‡	Pred. scale‡
1	0.15	0.79	+5.27	10.02	+++	+++
2	0.29	0.83	+2.86	9.94	+++	+++
3	0.31	0.49	+1.56	10.04	+++	+++
4	0.33	0.75	+2.27	9.99	+++	+++
5	0.46	0.98	+2.13	9.87	+++	+++
6	0.48	1.10	+2.29	9.80	+++	+++
7	0.55	0.50	−0.91	10.03	+++	+++
8	0.64	0.49	−0.76	9.75	+++	+++
9	0.69	1.10	+1.59	9.79	+++	+++
10	0.72	0.49	−0.68	9.95	+++	+++
11	0.76	0.49	−0.65	9.83	+++	+++
12	0.8	0.49	−0.61	10.08	+++	+++
13	1.1	0.89	−0.81	10.07	+++	+++
14	1.2	0.62	−0.52	9.76	+++	+++
15	1.3	0.90	−0.69	9.85	+++	+++
16	1.4	0.49	−0.35	9.89	+++	+++
17	1.4	2.40	+1.71	9.26	+++	++
18	2	1.01	−0.51	9.86	++	+++
19	2.1	3.20	+1.52	9.88	++	++
20	2.3	3.80	+1.65	9.75	++	++
21	2.5	1.80	−0.72	9.72	++	++
22	2.6	5.10	+1.96	9.57	++	++
23	2.7	1.90	−0.70	9.74	++	++
24	2.9	2.70	−0.93	9.43	++	++
25	3	2.59	−0.86	9.5	++	++
26	3.5	5.80	+1.66	9.02	++	++
27	3.8	4.60	+1.21	9.76	++	++
28	4.6	3.10	−0.67	9.67	++	++
29	10	5.12	−0.51	9.26	++	++
30	220	100	−0.45	5.73	++	++
31	670	5000	+7.46	5.26	+	+
32	16 000	1200	−0.08	7.52	+	+
33	100 000	110757	+1.11	7.03	+	+

*Difference between the predicted and experimental values. “+” indicates that the predicted IC₅₀ is higher than the experimental IC₅₀; “−” indicates that the predicted IC₅₀ is lower than the experimental IC₅₀; a value of 1 indicates that the predicted IC₅₀ is equal to the experimental IC₅₀.

†Fit value indicates how well the features in the pharmacophore overlap the chemical features in the molecule. Fit = weight × [max(0,1 –SSE)] where SSE = (D/T)², D is the displacement of the feature from the centre of the location constraints and T is the radius of the location constraint sphere for the feature (tolerance).

‡Activity scale: +++ (highly active) = IC₅₀ < 2 nM; ++ (moderately active) = 2 nM ≤ IC₅₀ ≤ 200 nM; + (low active) = IC₅₀ > 200 nM.

Figure 4. Best pharmacophore model Hypo1 aligned to a training set compound. (a) Active molecule compound 1 (IC₅₀: 0.15 nM) and (b) inactive molecule compound 33 (IC₅₀: 100 000 nM). Hydrogen bond acceptor (HBA, green), hydrogen bond donor (HBD, magenta), hydrophobic aromatic (Hy-Ar, blue) and ring aromatic (RA, brown).

Figure 5. The difference in costs between HypoGen runs and the scrambled runs. The 98% confidence level was selected.

Figure 6. 2D structure of the test set molecules used in the validation of Hypo1.

Table 4. Experimental and predicted IC₅₀ data values of 40 test set molecules against Hypo1.

Compound no.	Exp. IC50 nM	Pred. IC50 nM	Error*	Exp. scale†	Pred. scale†
1	0.77	1.2	+1.5	+++	+++
2	0.78	1.2	+1.5	+++	+++
3	0.95	0.79	−1.2	+++	+++
4	1.1	1.1	+1.0	+++	+++
5	1.1	0.76	−1.4	+++	+++
6	1.1	1.2	+1.1	+++	+++
7	1.1	0.64	−1.7	+++	+++
8	1.2	1.5	+1.3	+++	+++
9	1.2	0.65	−1.8	+++	+++
10	1.4	0.8	−1.7	+++	+++
11	1.4	0.6	−2.3	+++	+++
12	1.5	0.99	−1.5	+++	+++
13	1.5	0.71	−2.1	+++	+++
14	1.6	1.5	−1.1	+++	+++
15	1.7	0.77	−2.2	+++	+++
16	1.8	0.88	−2.1	+++	+++
17	1.9	0.62	−3.1	+++	+++
18	2	8.1	+4.0	++	++
19	2.9	8	+2.8	++	+
20	3	1.2	−2.5	++	++
21	3.3	0.86	−3.9	++	++
22	3.7	1.1	−3.2	++	++
23	3.8	1.1	−3.6	++	+++
24	3.8	8.4	+2.2	++	++
25	4	2.1	−1.9	++	++
26	4.2	0.79	−5.3	++	+++
27	4.3	2.2	−2.0	++	++
28	4.4	1.2	−3.6	++	+++
29	4.7	14	+2.9	++	++
30	4.7	0.71	−6.6	++	++
31	5.7	0.7	−8.2	++	+++
32	6.2	22	+3.6	++	+
33	7	0.83	−8.4	++	++
34	8.7	10	+1.2	++	++
35	10	25	+2.5	++	++
36	110	170	+1.6	++	++
37	110	170	+1.6	++	++
38	1100	5900	+5.3	+	+
39	1100	320	−3.4	+	+
40	1130	470	−2.4	+	+

*Difference between the predicted and experimental values. “+” indicates that the predicted IC₅₀ is higher than the experimental IC₅₀; “−” indicates that the predicted IC₅₀ is lower than the experimental IC₅₀; a value of 1 indicates that the predicted IC₅₀ is equal to the experimental IC₅₀.

†Activity scale: +++ (highly active) = IC₅₀ < 2 nM; ++ (moderately active) = 2 nM ≤ IC₅₀ ≤ 200 nM; + (low active) = IC₅₀ > 200 nM.

Table 5. Statistical parameter from screening test set molecules.

No.	Parameter	Values
1	Total number of molecules in database (D)	1300
2	Total number of actives in database (A)	15.00
3	Total number of hit molecules from the database (H^t)	18.00
4	Total number of active molecules in hit list (H^a)	15.00
5	Percentage of yield of actives [(H^a/H^t) × 100]	83.33
6	Percentage of ratio of actives [(H^a/A) × 100]	100
7	Enrichment factor (EF)	7.22
8	False negatives [A − H^a]	0
9	False positives [H^t − H^a]	3
10	Goodness of hit score* (GH)	0.87

*[(H_a/4H_tA) (3A + H_t) × (1 − ((H_t − H_a)/(D − A))].

Figure 7. Molecular-docking result. (a) NCI compound, (b) Chembridge compound, (c) Maybridge and (d) Maybridge hit compound bound in the active site of c-Src. Hydrogen bonds are shown in black.

Figure 8. Suitable molecular fragment for c-Src inhibitor obtained from the Bayesian model.

Figure 9. Non-suitable molecular fragment for c-Src inhibitor obtained from the Bayesian model.

Figure 10. Receiver operating characteristic (ROC) plot of Bayesian model for c-Src inhibitors.

Table 6. Statistical value obtained from the Bayesian model.

Leave-one out	XV ROC AUC	Best split			TP/FN FP/TN			# in Category
Bayesian model	0.940	7.059			17/1 1/11			18
Category statistics	Category N	Category Mean (±StdDev)			Non-category N			Non-category Mean (±StdDev)
Bayesian model	18	574.34 (±422.97)			12			−875.76 (± 1067.61)
Enrichment	Category %	1	5	10	25	50	75	90	95	99
Bayesian model	60%	5.6	11.1	22.2	38.9	83.3	100	100	100	100