Solubility is the amount of a substance that dissolves in a specified solvent under specified conditions. In drug discovery, it affects oral absorption, bioavailability, dosing, efficacy, and adverse-effect risk. Poor solubility can reduce absorption and require higher doses, while low oral bioavailability may also arise from permeability, metabolism, transporters, instability, or protein binding. section 4.0
AqSolDB makes solubility a regression problem: molecular structural features and chemical properties become inputs, and measured solubility becomes the target. Linear regression provides an interpretable baseline, with weights learned either through the normal equation or gradient descent. section 4.0 section 4.5
Practical modeling concerns include:
- Compare batch, mini-batch, and stochastic gradient descent while monitoring learning rate, convergence, gradient clipping, and feature scaling. section 4.5
- Evaluate regression performance and applicability domain, and inspect bias, variance, overfitting, and outlier sensitivity. section 4.0
- Use Scikit-Learn alternatives such as RANSAC for robust fitting and SVR for predictions within an epsilon margin. section 4.5
