QSPR flash point prediction of binary miscible mixtures: Integrated descriptors and DE-optimized SVR.

Monte Carlo optimization-based QSPR modeling of molar refractivity: Descriptor stability and applicability domain analysis.

An additive fragment-based QSPR model for molar refractivity (MR) was developed using Monte Carlo optimization within the CORAL framework. Molecular structure was encoded using fragment-level attributes, and their contributions were statistically optimized across multiple independent training-validation splits. The resulting linear models retain the additive character of classical fragment-constant approaches while introducing data-driven weight optimization under explicit statistical control. Across three independent splits, the developed QSPR models exhibit consistently high coefficients of determination in both the training and external validation sets, with validation R2 values ranging from approximately 0.86 to 0.89. Root-mean-square errors remained consistent across splits, indicating reproducible behavior of the optimized descriptors. An applicability domain (AD) was defined using statistical defect metrics derived from descriptor distributions. Across independent splits, the applicability domain consistently identified a minority of compounds with rare or unevenly represented SMILES attributes; these statistically under-supported compounds were associated with a modest increase in prediction error, indicating that the defect-based domain reflects descriptor representativeness rather than acting as a strict error filter. The study presents Monte Carlo-optimized additive modeling as a statistically audited extension of fragment-constant schemes, integrating robustness analysis and applicability-domain assessment into a transparent QSPR workflow.

Machine learning prediction of organic compound melting points informed by condensed-phase and electronic descriptors.

Melting point (MP) is an important thermophysical property for the chemical process industry, yet accurate prediction of MP for organic compounds in the absence of experimental data remains challenging due to the complex interplay between molecular packing, intermolecular interactions, and electronic structure. Traditional group contribution and quantitative structure-property relationship models, which rely primarily on static molecular descriptors, often fail to capture these critical condensed-phase effects. In this study, we present a hybrid machine learning framework that integrates cheminformatics descriptors with quantum chemical features and dynamic condensed-phase descriptors derived from molecular dynamics (MD) simulations. Using a curated subset of the DIPPR 801 database, multiple machine learning architectures, including light gradient boosting machine (LightGBM) and graph convolutional networks, were evaluated with feature sets of increasing physical fidelity. The best-performing model, based on LightGBM trained on Dragon descriptors augmented with MD and quantum chemical features, achieves a mean absolute error of 22.5 K, outperforming descriptor-only models and structure-based deep learning baselines. Shapley additive explanations interpretability analysis reveals that melting behavior is governed primarily by molecular topology, surface-area-weighted electronic descriptors, and condensed-phase interaction properties. In contrast, many isolated functional group and single molecule electronic descriptors contribute negligibly once these effects are accounted for. These results demonstrate that incorporating physics-informed, multi-scale descriptors enables more accurate and physically interpretable MP predictions.

Exploration and Screening of Database of Polycyclic Aromatic Hydrocarbons to Find Compounds with Lower Enthalpy of Formation

A Leakage-Aware Benchmark Study of Machine Learning Models for Deep Eutectic Solvent Property Prediction

This study provides a structured and reproducible assessment of the conditions under which descriptor-based ML models can be expected to succeed or fail in DES systems and highlights the importance of rigorous, leakage-aware evaluation in data-driven chemical modeling.

Machine learning guided discovery of deep eutectic solvents for NH 3 capture: Experimental validation and mechanism

Deep eutectic solvents (DESs) have considerable potential for NH 3 capture, but traditional solvent screening methods are unable to identify appropriate DES efficiently. One thousand nine hundred fifty‐nine experimental solubility data points for 72 DESs were used to construct and compare multiple machine learning models based on σ‐profile descriptors to predict NH 3 solubility in DESs. CatBoost achieved the best performance ( R 2  = 0.993, RMSE = 0.079). Nested cross‐validation and independent test sets confirmed that the model has good physical consistency and cross‐system generalization. SHAP analysis further quantified the contributions of key features. The final model was then employed to predict the NH 3 solubilities of 1140 DESs, and the highest‐ranked systems were selected for subsequent characterization and absorption experiments. The results of the gas absorption performance experiment are in excellent agreement with the predictions of the model. Finally, quantum chemical calculations were used to clarify the microscopic mechanisms underlying DES formation and NH 3 interaction.

Topological indices from M-polynomials for machine learning prediction of drug lipophilicity

Accurate prediction of physicochemical properties such as the octanol-water partition coefficient’s logarithm (LogP) is critical in early-stage drug development. This study presents a novel and interpretable computational framework that integrates symbolic graph-theoretical descriptors, specifically topological indices derived from M-polynomials, with machine learning (ML) techniques for LogP prediction in oncology drug molecules. Molecular graphs were constructed from SMILES strings, and eight M-polynomial-based topological indices were computed as predictive features. A comprehensive suite of regression models was applied, ranging from univariate and multivariate linear regressions to regularized methods (least absolute shrinkage and selection operator (LASSO), ridge, ElasticNet) and nonlinear ensemble learners (random forest, XGBoost). Dimensionality reduction using principal component analysis (PCA) and rigorous validation via cross-validation and bootstrapping were conducted. Among all approaches, ensemble models combining XGBoost and random forest with bootstrapping yielded the most robust performance, achieving an R2 score greater than 0.6 and a low prediction error. These results demonstrate the effectiveness of M-polynomial-derived indices as interpretable molecular descriptors and affirm the predictive utility of advanced ML models in quantitative structure-property relationship (QSPR) modeling. To our knowledge, this is among the first comprehensive studies integrating ensemble-based ML methods and M-polynomial-derived topological indices for LogP prediction in oncology drugs.