Predicting the solubility and lipophilicity of platinum(II, IV) complexes is essential for prioritizing potential anticancer candidates in drug discovery. This study introduces the first publicly available online model for predicting the solubility of platinum complexes, addressing the lack of literature and models in this regard. Using a time-split dataset, we developed a consensus model with a Root Mean Squared Error (RMSE) of 0.62 through 5-cross-validation on a training set of 284 historical compounds (solubility data reported prior to 2017). However, the RMSE increased to 0.86 when applied to a prospective test set of 108 compounds reported after 2017. Further analysis of the high prediction errors revealed that these inaccuracies are primarily attributed to the underrepresentation of novel chemical scaffolds, particularly Pt(IV) derivatives, in the training sets. For instance, a series of eight phenanthroline-containing compounds, not covered by the training set's chemical space, had an RMSE of 1.3. When the model was redeveloped using a combined dataset, the RMSE of this series significantly decreased to 0.34 under the same validation protocol. Additionally, we developed an interpretable linear model to identify structural features and functional groups that influence the solubility of platinum complexes. We further validated the correlation between solubility and lipophilicity, consistent with the Yalkowsky General Solubility Equation. Building on these insights, we developed a final multitask model that simultaneously predicts solubility and lipophilicity as two endpoints with RMSE = 0.62 and 0.44, respectively. The data and final developed model is available at https://ochem.eu/article/31.
