Candidate Inhibitor Discovery by Coupling Machine Learning with Physics-Based Models
Graphical abstract
Summary
Ligand-feature learning, structure-based screening, dynamics and property assessment are combined into a traceable prioritisation path. The graphical abstract brings together candidate distributions, structural review and property comparisons across the screening funnel. The results establish a clear candidate hierarchy for experimental selection and subsequent lead optimisation. The result-focused presentation supports efficient review of the main evidence and research priorities.
Computational results
Data distribution and model training
Figure 1. Distribution of activity data in training and testing centers.
This figure presents the principal structures and trends in “Data distribution and model training” and connects them to the case-level ranking and result interpretation.
Figure 2. Machine learning is the process of learning.
This figure presents the principal structures and trends in “Data distribution and model training” and connects them to the case-level ranking and result interpretation.
Model testing and selection
Figure 3 - The remaining graphs based on the test set.
This figure presents the principal structures and trends in “Model testing and selection” and connects them to the case-level ranking and result interpretation.
Scroll horizontally when the table is wider than the page
| Model name | MSE | R² |
|---|---|---|
| LightGBM | 0.298 | 0.774 |
| Random forests | 0.323 | 0.755 |
| Type R2 | 0.370 | 0.719 |
| Growth gradually | 0.372 | 0.716 |
| Multilayered Perceptron | 0.436 | 0.669 |
| The back of the spine | 0.492 | 0.627 |
| The Tree of Decision | 0.572 | 0.566 |
| Nearby neighbors | 0.786 | 0.403 |
| Supporting vectors | 1.150 | 0.128 |
| The powerful net is back. | 1.196 | 0.093 |
| The return of Laso | 1.197 | 0.092 |
| Linear regression | 1.84E+17 | -1.40E+17 |
This table consolidates the key comparisons in “Model testing and selection”, making differences across conditions and candidates directly reviewable.
Full case PDF
Complete results, high-resolution figures and analysis are provided in the client PDF. Add our service account on WeChat and send HS-CASE-0032 to request it.