Themed collection 2025 Digital Discovery Emerging Investigators

A scalable and practical protocol for estimating the expectation values of fermionic observables is presented. The approach is based on an iterative procedure that measures low-cost operator groups across different orbital bases.

Correcting the Nernst–Einstein equation using sigma profile-based machine learning models is an accurate, interpretable approach to estimate ionic liquid conductivities.

Pessimistic model predictions in asynchronous Bayesian optimization can enable more efficient and robust experimental system optimization in both asychronous and serial sampling settings.

FiberForge provides an end-to-end platform for molecular modeling and design of amyloid materials, enabling physics-based identification of sequences and polymorphs with targeted mechanical behavior.

Determination of vibrational modes in aromatic VOCs via DFT and Bayesian inference to match theoretical and experimental spectra.

In this report, the compounds listed in DrugBank were structurally mapped to a commercial catalog of chemical feedstocks through reaction agnostic one step retrosynthetic disconnection.

Fullerenes are sp²-carbon carges with extensive isomeric diversity. A deep learning model is developed to accurately predict their stability, solubility, and electronic properties by learning directly from topological features in the structures.

Automated active learning integrated with enhanced sampling facilitates data-efficient training of machine learning interatomic potentials for chemical reactions.

ReactPyR enables automated ReactIR workflows to quantify air-sensitivity in organometallic reagents, delivering reproducible kinetic insights and guiding stabilisation strategies for safer, more efficient handling of highly reactive species.

When data are scarce, surrogate models trained on QM descriptors help. We report that their hidden features, not the predicted descriptors, provide superior inputs for downstream predictive chemistry tasks.

The accurate prediction of reaction barrier heights is crucial for understanding chemical reactivity and guiding reaction design.

We propose novel SMILES augmentation strategies for chemical language modelling, which broaden the range of tools for generative drug design.

Combinatorial liquid handling can be time-consuming. We reframe it as a capacitated vehicle routing problem (CVRP), enabling up to a 37% reduction in execution time across diverse labware formats using a heuristic solver.

Parallel, inhomogeneous and inherently stochastic, the aerosol medium holds exceptional promise in the unfolding era of digitisation as a platform for synthesis and discovery tailored to programmable execution and rapid computational analysis.

We introduce an activity-cliff explanation supervision training strategy to enhance both predictivity and explainability for graph neural networks in molecular structure and activity relationship modeling.

We present a self-driving fluidic lab with a modular hardware and software for data-driven synthesis optimization of eco-friendly colloidal semiconductor nanocrystals.