Table 5.
Comparison of generated ligands for ESR1 and ACAA1 following multi-objective optimization and refinement.
| Ligand | Optimized prop. | Non-optimized prop. | ||||||
|---|---|---|---|---|---|---|---|---|
| KD (AD) (↓) | QED (↑) | SA (↓) | KD (ABFE) (↓) | Lipinski | PAINS (↓) | Fsp3 (↑) | MCE-18 (↑) | |
|
| ||||||||
| ESR1 | ||||||||
|
| ||||||||
| LIMO mol. #1 | 4.6 | 0.43 | 4.8 | 6 · 10−5 | ✔ | 0 | 0.16 | 90 |
| LIMO mol. #2 | 2.8 | 0.64 | 4.9 | 1000 | ✔ | 0 | 0.52 | 76 |
|
| ||||||||
| GCPN mol. #1 | 810 | 0.43 | 4.2 | - | ✔ | 0 | 0.29 | 22 |
| GCPN mol. #2 | 2.7 · 104 | 0.80 | 3.7 | - | ✔ | 0 | 0.56 | 47 |
|
| ||||||||
| Tamoxifen | 87 | 0.45 | 2.0 | 1.5* | ✔ | 0 | 0.23 | 16 |
| Raloxifene | 7.9 · 106 | 0.32 | 2.4 | 0.030* | ✔ | 0 | 0.25 | 59 |
|
| ||||||||
| ACAA1 | ||||||||
|
| ||||||||
| LIMO mol. #1 | 28 | 0.57 | 5.5 | 4 · 104 | ✔ | 0 | 0.52 | 52 |
| LIMO mol. #2 | 31 | 0.44 | 4.9 | No binding | ✔ | 0 | 0.81 | 45 |
|
| ||||||||
| GCPN mol. #1 | 8500 | 0.69 | 4.2 | - | ✔ | 0 | 0.52 | 61 |
| GCPN mol. #2 | 8500 | 0.54 | 4.3 | - | ✔ | 0 | 0.52 | 30 |
Arrows indicate whether a high score (↑) or low score (↓) is desired. High QED, Fsp3, and satisfying Lipinski’s Rule of 5 suggest drug-likeness. A low number of PAINS alerts indicates a low likelihood of false positive results in binding assays. MCE-18 is a measure of molecular novelty based on complexity, and SA is a measure of synthesizability. KD values in nM are computed binding affinities from AutoDock-GPU (AD) and from more rigorous absolute binding free energy calculations (ABFE). See Appendix A.2 for a full description of each metric.
*
indicates an experimentally determined value obtained from BindingDB (Liu et al., 2007).