When Do Simple Models Win? Machine Learning Architectures for UV Absorption Prediction and Insights for General Molecular Property Prediction

What determines which machine learning architecture wins for molecular property prediction? We address this question through a controlled comparison of five models spanning four architecture families: fingerprint-based ensembles (Random Forest, XGBoost), a directed message-passing graph neural network (GNN), a bidirectional gated recurrent network (BiGRU), and a pretrained Transformer, using UV absorption wavelength (λ max) prediction as a testbed across 18,755 solute-solvent pairs with stratified 5-fold cross-validation and two external datasets totaling over 40,000 molecules. The answer depends on the task. For screening within known chemical space, RF with Morgan fingerprints is optimal: it matches the GNN (RMSE = 31.34 ± 1.82 vs. 31.69 ± 3.18 nm, p = 0.77), trains in 15 minutes on a CPU with no hyperparameter tuning, and outperforms all sequence-based models. For exploring novel scaffolds, deep learning is superior: experimental validation on 16 novel UV-absorbing 1 compounds shows that the GNN (MAE = 26.2 nm), Transformer (26.3 nm), and Bi-GRU (28.6 nm) all outperform RF (38.5 nm), as learned representations generalize where fixed fingerprints cannot. In either case, encoding solvent identity improves all models: a simple SMILES/fingerprint concatenation strategy yields 6-8% RMSE reductions across fingerprint, graph, and recurrent architectures (p < 0.02), without domain-specific descriptor engineering. Interpretability analysis confirms that both RF feature importance and BiGRU gradient saliency independently highlight the same chromophore motifs, and that architectures with local inductive bias systematically outperform global-attention Transformers on this locally determined property. These findings provide general guidance for molecular property prediction: the largest or most complex model is not always the best choice. Instead, optimal model selection depends on the locality of the target property and the relationship between target compounds and available training data. When the target property is governed by local molecular features and the prediction task is interpolation within known chemical space, simple fingerprint-based models offer an efficient and accurate alternative to deep learning.