Worked Examples

Four drug targets · docking rescoring · RPS uncertainty analysis

Each example shows docking scores alongside single-frame MM/GBSA rescores, the resulting rank changes, and Rapid Perturbation Sampling (RPS) statistics. All MM/GBSA scores are reported in kcal/mol. A negative ΔRank means the compound moved up in ranking (more favoured) after rescoring; a positive ΔRank means it moved down.

ΔRank negative – compound improved in MM/GBSA ranking

ΔRank positive – compound dropped in MM/GBSA ranking

CV > 15% – score sensitive to coordinate perturbation

EGFR Kinase Inhibitors

Empirical docking scores for the EGFR ligand set favoured bulkier and more hydrophobic compounds, producing a ranking largely driven by steric complementarity. Application of Pose-Rescorer gave a reordered ligand ranking based on single-frame MM/GBSA scores.

Table 1: Docking scores and single-frame MM/GBSA rescoring for EGFR inhibitors

Ligand	Docking (kcal/mol)	Rank	Rescore (kcal/mol)	Rank	ΔRank
Lapatinib	−10.63	1	−96.06	3	+2
TAK-285	−10.02	2	−100.33	1	−1
AEE788	−9.82	3	−78.97	4	+1
Gefitinib	−9.40	4	−99.10	2	−2
Icotinib	−9.22	5	−21.62	10	+5
PKI-166	−9.19	6	−60.55	7	+1
Vandetanib	−9.10	7	−74.24	6	−1
AG1478	−8.39	8	−42.18	9	+1
PD153035	−8.32	9	−43.02	8	−1
Erlotinib	−8.01	10	−75.24	5	−5

Table 2: RPS summary statistics for EGFR inhibitors (σ = 0.20 Å, 10 replicates)

Ligand	Mean (kcal/mol)	SD	CV (%)
Gefitinib	−92.62	3.39	3.66
Lapatinib	−95.46	4.85	5.08
TAK-285	−97.09	6.80	7.01
AEE788	−95.18	7.00	7.35
Erlotinib	−77.78	6.18	7.95
Vandetanib	−77.69	5.71	7.35
PKI-166	−57.37	4.32	7.53
AG1478	−49.86	5.52	11.07
PD153035	−40.87	4.46	10.92
Icotinib	−23.80	7.40	31.07

Icotinib ranks last by MM/GBSA (−21.62 kcal/mol) and exhibits a high CV of 31.07%, indicating that its score is highly sensitive to small coordinate perturbations and should be interpreted with caution. TAK-285 and Gefitinib, which rise in rank after rescoring, show low CVs (<8%), confirming their scores are geometrically stable.

HIV-1 Protease Inhibitors

Docking of the HIV-1 protease inhibitor set produced a ranking strongly influenced by steric complementarity within the compact active-site cavity. Single-frame MM/GBSA rescoring resulted in substantial reordering of ligands.

Table 3: Docking and rescoring for HIV-1 protease inhibitors

Ligand	Docking (kcal/mol)	Rank	Rescore (kcal/mol)	Rank	ΔRank
Saquinavir	−10.97	1	−86.97	1	0
Indinavir	−10.69	2	−81.35	2	0
Lopinavir	−10.44	3	−72.22	4	+1
Tipranavir	−10.00	4	−54.37	9	+5
Atazanavir	−9.74	5	−75.30	3	−2
Brecanavir	−9.65	6	−58.29	8	+2
Nelfinavir	−9.54	7	−61.42	7	0
Ritonavir	−9.40	8	−66.80	6	−2
Darunavir	−9.04	9	−65.31	5	−4
Amprenavir	−8.71	10	−54.74	10	0

Table 4: RPS summary statistics for HIV-1 protease inhibitors (σ = 0.20 Å, 10 replicates)

Ligand	Mean (kcal/mol)	SD	CV (%)
Atazanavir	−80.07	2.70	3.37
Amprenavir	−54.02	3.26	6.04
Ritonavir	−88.08	5.54	6.29
Nelfinavir	−58.80	4.16	7.07
Darunavir	−64.82	4.66	7.19
Indinavir	−76.68	5.67	7.39
Lopinavir	−75.51	6.96	9.21
Tipranavir	−52.84	5.76	10.89
Brecanavir	−58.22	7.65	13.13
Saquinavir	−63.87	8.43	13.20

Tipranavir drops five positions after rescoring (rank 4 → 9) despite a docking score of −10.00 kcal/mol, consistent with known difficulties in scoring its unusual non-peptidic scaffold. All RPS CVs remain below 15%, indicating the entire set has moderate-to-high score stability. Brecanavir and Saquinavir show the highest CVs (13.1–13.2%), approaching the MODERATE/LOW boundary.

BRD4 Bromodomain Inhibitors

Docking of the BRD4 inhibitor set produced a ranking influenced primarily by steric complementarity within the shallow acetyl-lysine recognition pocket. Single-frame MM/GBSA rescoring resulted in pronounced reordering of ligands.

Table 5: Docking and rescoring for BRD4 inhibitors

Ligand	Docking (kcal/mol)	Rank	Rescore (kcal/mol)	Rank	ΔRank
CPI-0610	−9.95	1	−23.11	9	+8
Molibresib	−9.49	2	−36.01	6	+4
Birabresib	−9.42	3	−38.47	3	0
Mivebresib	−9.02	4	−32.03	8	+4
JQ1	−8.65	5	−41.21	1	−4
Bromosporine	−8.40	6	−36.85	5	−1
MS436	−8.00	7	−37.82	4	−3
I-BET151	−7.98	8	−40.53	2	−6
PFI-1	−7.52	9	−33.40	7	−2
Apabetalone	−7.30	10	−22.15	10	0

Table 6: RPS statistics for BRD4 inhibitors (σ = 0.20 Å, 10 replicates)

Ligand	Mean (kcal/mol)	SD	CV (%)
Molibresib	−35.67	1.05	2.94
MS436	−37.38	1.86	4.98
JQ1	−41.18	2.19	5.31
Birabresib	−39.82	2.24	5.62
Apabetalone	−26.32	1.76	6.69
PFI-1	−31.19	2.16	6.92
Mivebresib	−38.37	2.49	6.49
I-BET151	−42.53	2.76	6.49
CPI-0610	−27.64	2.35	8.50
Bromosporine	−33.96	3.58	10.54

BRD4 shows the most dramatic rank inversions in this benchmark: CPI-0610 drops eight positions (rank 1 → 9) and I-BET151 rises six positions (rank 8 → 2). The overall score magnitudes are smaller than EGFR/HIV-1 protease, reflecting the shallower BRD4 pocket. All RPS CVs remain below 11%, confirming score stability across the set.

HSP90 Chaperone Inhibitors

Docking of the HSP90 inhibitor set gave a ranking dominated by steric complementarity within the N-terminal ATP-binding pocket, favouring larger ligands capable of spanning the elongated binding channel. Substantial reordering occurred upon rescoring.

Table 7: Docking and rescoring for HSP90 inhibitors

Ligand	Docking (kcal/mol)	Rank	Rescore (kcal/mol)	Rank	ΔRank
SNX-2112	−12.06	1	−69.22	1	0
Ganetespib	−10.24	2	−43.56	6	+4
Luminespib	−9.53	3	−58.92	3	0
AT13387	−9.39	4	−64.94	2	−2
BIIB021	−8.46	5	−44.32	5	0
Radicicol	−8.31	6	−25.54	10	+4
PU-H71	−8.31	7	−54.76	4	−3
Alvespimycin	−8.02	8	−42.47	7	−1
Geldanamycin	−7.90	9	−33.98	8	−1
Tanespimycin	−7.76	10	−33.73	9	−1

Table 8: RPS statistics for HSP90 inhibitors (σ = 0.20 Å, 10 replicates)

Ligand	Mean (kcal/mol)	SD	CV (%)
Luminespib	−56.66	2.82	4.98
PU-H71	−60.88	3.06	5.03
Radicicol	−63.11	3.61	5.72
AT13387	−58.41	3.61	6.18
SNX-2112	−54.92	3.80	6.92
BIIB021	−65.35	4.60	7.04
Tanespimycin	−60.88	4.44	7.29
Ganetespib	−65.31	6.06	9.28
Alvespimycin	−70.01	6.85	9.78
Geldanamycin	−60.88	6.85	11.25

Radicicol drops four positions after rescoring (rank 6 → 10), consistent with its natural product scaffold having atypical electrostatic interactions that docking scores overestimate. PU-H71 rises three positions (rank 7 → 4), benefitting from improved electrostatic treatment under GB solvation. All RPS CVs remain below 12%, indicating good numerical stability across the HSP90 set.

Cross-Target Summary

Across all four targets, single-frame MM/GBSA rescoring consistently reorders ligands relative to docking. The degree of reordering varies by target, reflecting the interplay between pocket character (electrostatic vs. hydrophobic) and scoring function approximations.

Target	Ligands	Max \|ΔRank\|	Mean CV (%)	CV > 15%
EGFR	10	5	7.9	1 (Icotinib, 31.1%)
HIV-1 Protease	10	5	8.4	0
BRD4	10	8	6.7	0
HSP90	10	4	7.6	0

Interpretation reminder: MM/GBSA scores are relative and rank-oriented. Absolute values should not be compared across different protein targets or interpreted as binding free energies. RPS CVs quantify numerical sensitivity to coordinate uncertainty only — they do not reflect conformational or thermodynamic uncertainty.