Correlation coefficient values of best ML-QSAR regression models.

ML method	Selected model descriptorsa,b,c	r 2 d	r L20% 2 e	r LOO 2 f	r PRESS 2 g
GFA-SVR	Hypo(5-R2-08), Hypo(6-R2-07), Hypo(8-R3-08), LEU244HNLD, VAL324HBLD, ASP325HALD, CHI_2, Num_Rings6	0.91	0.65	0.66	0.76
GFA-RF	Hypo(5-R2-08), Hypo(6-R2-03), Hypo(8-R3-08), Hypo(2-R5-05), VAL324HBLD, ASP325HALD, CHI_2	0.94	0.63	0.57	0.77
GFA-PNN	Hypo(3-R6-08), Hypo(1-R6-02), LEU210HD12CD, Num_Rings5, Kappa_3	0.96	0.07	0.01	0.71
GFA-XGBoost	Hypo(5-R2-07), Hypo(6-R2-08), Hypo(8-R2-04), Hypo(2-R5-05), Kappa_3, Dipole_Y	0.96	0.47	0.46	0.75
GFA-MLR	log(1/IC50) = + 0.12 Hypo(5-R6-08) + 0.129 Hypo(1-R2-08) − 0.276 LYS191HZ2CD − 0.22 VAL324 HBLD + 0.433 Num_Rings5 − 0.002 PMI_x − 2.65 Shadow_XYfrac − 1.667	0.65	0.46	0.50	0.53

^aHypo(5-R2-08) is the 8^th pharmacophore model generated using training subset 5 (Table S2 under ESI) with the 2^nd HYPOGEN run settings (Table S3 under ESI), Hypo(6-R2-07): is the7^th pharmacophore model generated using training subset 6 (Table S2) with the 2^nd HYPOGEN run settings (Table S3), Hypo(8-R3-08) is the 8^th pharmacophore model generated using training subset 8 (Table S2) with the 3^rd HYPOGEN run settings (Table S3), Hypo(6-R2-03) is the 3^rd pharmacophore model generated using training subset 6 (Table S2) with the 2^nd HYPOGEN run settings (Table S3), Hypo(2-R5-05) is the 5^th pharmacophore model generated using training subset 2 (Table S2) with the 5^th HYPOGEN run settings (Table S3), Hypo(3-R6-08) is the 8^th pharmacophore model generated using training subset 3 (Table S2) with the 6^th HYPOGEN run settings (Table S3), Hypo(1-R6-02) is the 2^nd pharmacophore model generated using training subset 1 (Table S2) with the 6^th HYPOGEN run settings (Table S3), Hypo(5-R2-07) is the 7^th pharmacophore model generated using training subset 5 (Table S2) with the 2^nd HYPOGEN run settings (Table S3), Hypo(6-R2-08) is the 8^th pharmacophore model generated using training subset 6 (Table S2) with the 2^nd HYPOGEN run settings (Table S3), Hypo(8-R2-04) is the 4^th pharmacophore model generated using training subset 8 (Table S2) with the 2^nd HYPOGEN run settings (Table S3), Hypo(5-R6-08) is the 8^th pharmacophore model generated using training subset 5 (Table S2) with the 6^th HYPOGEN run settings (Table S3), Hypo(1-R2-08) is the 8^th pharmacophore model generated using training subset 1 (Table S2) with the 2^nd HYPOGEN run settings (Table S3). Table 2 shows the X, Y, Z coordinates of pharmacophores Hypo(5-R2-08), Hypo(6-R2-07), and Hypo(8-R3-08).

^bLEU244HNLD is the hydrogen atom attached to peptidic N of Leu244 selected by LibDock score scoring function, VAL₃₂₄HB^LD is the hydrogen atom attached to beta carbon of Val324 selected by LibDock score scoring function, ASP₃₂₅HA^LD is the hydrogen atom attached to alpha carbon of Asp325 selected by LibDock score scoring function. Fig. 3 shows the position of these three atoms within the binding pocket, LEU₂₁₀HD12^CD is the hydrogen atom attached to delta carbon of Leu210 selected by CDocker interaction energy scoring function, LYS₁₉₁HZ2^CD is one of the hydrogen atoms at the terminal amine on the side chain of Lys191 selected by CDocker interaction energy scoring function.

^cNum_Rings6: number of 6-membered rings. CHI_2: second order connectivity index, positively correlated with molecular size, Num_Rings5: number of 5-membered rings. Kappa_3: third order kappa shape index, related to molecular flexibility, Dipole_Y: 3D the calculated magnitude and the X-vector component of the molecular dipole moment in debyes as estimated from the partial atomic charges (calculated by Gasteiger method) and atomic coordinates. PMI_x: principle moment of inertia in the X-dimension, Shadow_XYfrac area of the molecular shadow in the XY plane.^54,55

^dResubstitution correlation coefficient: the model is trained on the training list and used to predict the bioactivities of the same training set.

^eLeave-20%-out correlation coefficient.

^fLeave-one-out correlation coefficient.

^gPredictive correlation coefficient on the external testing set.