A convex hull is the smallest convex region containing all training points in descriptor space. For molecular solubility models, it provides a more precise applicability domain than descriptor-wise minimum and maximum bounds. section 4.1.3
Use the convex hull to:
- Assess whether a new molecule lies within the region represented by the training data. section 4.1.3
- Avoid some empty regions that a bounding box falsely labels as reliable and account for descriptor relationships. section 4.1.3
- Recognize that computation becomes unwieldy as descriptor dimensionality increases and that the method retains some bounding-box limitations. section 4.1.3
Figure 4.6 Estimated model AD based off of regions defined by the bounding box (dashed line) and convex hull (dotted line) methods.

Compare the convex-hull applicability domain with the broader bounding-box region in descriptor space. section 4.1.3




