Reviews & Analysis

Digital twins of self-driving chemistry laboratories may help reduce reliance on costly real-world experimentation and enable the testing of hypothetical automated workflows in silico.

This Review explores the integration of deep learning in first-principles electronic structure calculations, addressing the accuracy–efficiency dilemma of traditional algorithms and extending first-principles methods to unprecedented scales and complexity.

Quantum error correction is vital for scalable quantum computing, but it incurs high resource overheads. This Perspective outlines recent breakthroughs and explores the opportunities to reduce the overheads by co-designing across algorithms, error-correction schemes and hardware architecture.

As quantum mechanics marks its centennial, machine learning interatomic potentials are emerging as transformative tools bridging quantum accuracy with classical efficiency. This Perspective explores their evolution in terms of accuracy, efficiency, interpretability and generalizability challenges.

SynGFN integrates synthesis constraints directly into the chemical design process. The result is a generative framework that produces diverse, high-quality molecules that can be readily synthesized in the laboratory.

A recent study demonstrates the applicability of quantum computers for multi-objective optimization, bringing quantum computing a step closer towards practical applications.

Scouter, a deep learning approach, predicts transcriptional responses to genetic perturbations by integrating large language model (LLM)-based gene embeddings with a lightweight compressor–generator neural network, providing valuable insights into the application of LLMs to biological research.

We developed an open-source, prototype AI collaborator for the science of science (SciSci). Through a web-based chat interface, SciSciGPT orchestrates auditable, automated workflows for literature understanding and data processing, analytics, and visualization. The system accelerates early-stage idea exploration, prototyping, and iteration, while improving reproducibility and accessibility for SciSci researchers.

A physics-infused heterogeneous graph neural network has been developed to address challenges in designing complex nanomaterials with spatially varying compositions. This fully differentiable model enables the rapid optimization and discovery of photon upconverting nanoparticle heterostructures that are 6.5-fold brighter than any nanoparticle in the training set.

A recent study demonstrates the efficiency of quantum-mechanical modeling of material properties by mapping the problem onto neuromorphic device architectures.

Mass spectrometry data analysis has long been limited to known molecules and exact matches. In a recent manuscript, a scalable search algorithm is proposed for uncovering both known compounds and novel molecular variants, enabling insights into natural product biosynthesis.

A framework called AUTOENCODIX benchmarks diverse autoencoder architectures in biological molecular profiling data, enabling insights from complex, multi-layered data.

Quantum computers are inching closer to practical deployment, but shielding fragile quantum information from errors is still very challenging. Now, a machine-learning-based decoder offers a strategy for rectifying errors in logic quantum circuits, hastening the advent of reliable and fault-tolerant quantum systems.

A recent study proposes efficient numerical algorithms to reduce the required computational resources for solving the edge states in large-scale photonic or acoustic structures.

Research now suggests that large language models (LLMs) are viable in silico models of human language processing. By examining multi-participant high-quality brain responses, researchers were able to break new ground in the validation of this proposal, which could dramatically reduce the barrier to studying how language is processed in the human brain.

A systematic comparison of large language models suggests that larger models align better with both human behavior and brain activity during natural reading. Instruction tuning, however, does not yield a similar benefit.

A recent study highlights how data changes not only how we can assess the performance of legal firms in the US, but more broadly how computational science is expanding beyond its traditional scope and into the legal field.

A recent study proposes using a single neural network to model and compute a wide range of solid-state materials, demonstrating exceptional transferability and substantially reduced computational costs — a breakthrough that could accelerate the design of next-generation materials in applications from efficient solar cells to room-temperature superconductors.

The Large Perturbation Model (LPM) is a computational deep learning framework that predicts gene expression responses to chemical and genetic perturbations across diverse contexts. By modeling perturbation, readout, and context jointly, LPM enables in silico hypothesis generation and drug repurposing.

The recent computational model ‘BRyBI’ proposes that gamma, theta, and delta neural oscillations can guide the process of word recognition by providing temporal windows for the integration of bottom-up input with top-down information.