Decoding the chemical space of fast-ion conductors via a descriptor-guided transfer learning framework

End-to-End Deep Learning Models for Predicting the Electrical Conductivity of Ionic Liquids

Ionic liquids (ILs) hold great promise for electrochemical technologies, yet their virtually infinite compositional space renders exhaustive experimental conductivity measurements infeasible. Existing QSPR models frequently suffer from limited generalizability when evaluated under strict IL-disjoint partitioning, and those that perform reasonably often depend on costly COSMO-RS computations. Here we present end-to-end multilayer perceptron (MLP) models that directly forecast IL electrical conductivity from a minimal set of readily accessible molecular features—temperature, Dragon structural descriptors, and quantum chemical parameters—entirely circumventing continuum solvation calculations. Two curated data sets were examined (Data Set I: 2168 records/242 ILs; Data Set II: 5297 records/629 ILs), each partitioned via rigorous IL-based splitting such that test sets comprised 48 and 157 completely unseen ILs, respectively. Our MLP architectures attained a test coefficient of determination (R2) of 0.871 with a mean absolute error (MAE) of 0.392 on Data Set I and R2 of 0.778 (MAE = 0.600) on Data Set II, on par with COSMO-RS-boosted alternatives while eliminating their computational overhead and convergence issues. A random forest control experiment further demonstrated that point-wise splitting inflates R2 to 0.979, whereas the identical model collapses to 0.627 under IL-disjoint validation, highlighting the critical role of proper data partitioning. SHAP-based interpretability pinpointed temperature, molecular geometry, polarizability distribution, and nitrogen-bearing fragments as dominant conductivity modulators. This study furnishes an efficient, robust, and COSMO-RS-independent computational toolkit for high-throughput virtual screening of conductive IL candidates.

Rapid Generative Discovery of High‐Energy Molecules in Low Data Regimes Using Minimal Computational Resources

This work presents a novel approach toward high‐energy molecules by combining long short‐term memory (LSTM) networks for molecular generation and attentive graph neural networks (GNN) for property predictions by combining fixed SHA‐256 embeddings with partially trainable representations.

GFN1-xTB-Assisted Machine Learning for Electronic Structure Screening of Metal–Organic Frameworks

A semiempirical extended tight-binding approach (GFN1-xTB) is employed to compute the electronic properties of a dataset of MOFs, and it is shown that GFN1-xTB approximates MOF band gaps well, as compared to semilocal DFT.

Stoichiometric cluster learning for few-shot property prediction of multi-ionic integrated energetic materials

It is shown how pretrained machine-learned interatomic potentials (MLIPs) can bypass full crystal-structure prediction and support pre-synthesis screening from stoichiometric ionic clusters using multi-ionic integrated explosives (MIXs) as a synthesis-facing example.

Generalized Global Neural Network with Pairwise Charge Transfer for Fast Prediction of Dynamic Properties under an Electric Field.

Machine learning potentials (MLPs) have emerged as game-changing tools for large-scale atomic simulations, overcoming the poor-scaling limitation intrinsic to traditional quantum mechanics (QM) methods. However, accurately incorporating electronic information remains a significant challenge for MLPs, particularly in efficiently computing the dynamic properties of matter under electric fields─a task central to topics such as infrared spectroscopy, interfaces under electric fields, and ferroelectric polarization. Herein, we report a physics-informed pairwise charge-transfer (PQT) theory to derive dynamic equations for macroscopic polarization that inherently conserve fundamental physical laws. Using the PQT theory, a Generalized Global Neural Network (GGNN) enhanced with the PQT mechanism is developed for the rapid prediction of dynamic properties under electric fields, applicable to both molecules and materials across the periodic table. Specifically, a generalized global data set comprising 3.18 million structures with atomic charges for 81 elements is utilized to pretrain a GGNN patched with PQT modules. Leveraging this pretrained GGNN-PQT potential, we can conveniently sample the potential energy surface under electric fields and fine-tune the potential using a small QM data set containing exact response properties at low cost. Our GGNN-PQT has linear scaling and introduces a low computational overhead compared to the standard GGNN, yet achieves both high speeds and low scaling. We demonstrate the performance of GGNN-PQT in computing dynamic response properties across a wide range of systems, including isolated molecules, adsorbed molecules, molecular crystals, liquid water, and ferroelectric materials.

Stability prediction of organic field-effect transistors by machine learning