Articles in 2024

Previous studies have explored the integration of episodic memory into reinforcement learning and control. Inspired by hippocampal memory, Freire et al. develop a model that improves learning speed and stability by storing experiences as sequences, demonstrating resilience and efficiency under memory constraints.

Liu et al. developed a framework called ARNLE to explore host tropism of SARS-CoV-2 and found a shift from weak to strong primate tropism. Key mutations involved in this shift can be analysed to advance research on emerging viruses.

Self-supervised learning techniques are powerful assets for enabling deep insights into complex, unlabelled single-cell genomic data. Richter et al. here benchmark the applicability of self-supervised architectures into key downstream representation learning scenarios.

Matthews et al. present a protein sequence embedding based on data from ancestral sequences that allows machine learning to be used for tasks where training data are scarce or expensive.

A kernel approximation method that enables linear-complexity attention computation via analogue in-memory computing (AIMC) to deliver superior energy efficiency is demonstrated on a multicore AIMC chip.

Survival prediction models used in healthcare usually assume that training and test data share a similar distribution, which is not true in real-world settings. Cui and colleagues develop a stable Cox regression model that can identify stable variables for predicting survival outcomes under distribution shifts.

Approaches are needed to explore regulatory RNA motifs in plants. An interpretable RNA foundation model is developed, trained on thousands of plant transcriptomes, which achieves superior performance in plant RNA biology tasks and enables the discovery of functional RNA sequence and structure motifs across transcriptomes.

Ektefaie and colleagues introduce the spectral framework for models evaluation (SPECTRA) to measure the generalizability of machine learning models for molecular sequences.

Reconstructing and predicting spatiotemporal dynamics from sparse sensor data is challenging, especially with limited sensors. Li et al. address this by using self-supervised pretraining of a generative model, improving accuracy and generalization.

Interactive robots can be used to study animal social behaviour. Imitation learning can be used to enable a rat-like robot to learn subtle templates of social behaviour, demonstrating that it can modulate the emotional states of rats through varied interaction patterns.

Predicting nanobody–antigen interactions is crucial for advancing nanobody development in drug discovery, but it remains a challenging task. Deng et al. propose DeepNano to enhance the prediction of nanobody–antigen interactions, facilitating virtual screening of target nanobodies.

Rate- and noise-induced transitions pose key tipping risks for ecosystems and climate subsystems, yet no predictive theory existed before. This study introduces deep learning as an effective prediction tool for these tipping events.

Ock and colleagues explore predictive and generative language models for improving adsorption energy prediction in catalysis without relying on exact atomic positions. The method involves aligning a language model’s latent space with graph neural networks using graph-assisted pretraining.

Why brain-like feature extraction emerges in large language models (LLMs) remains elusive. Mischler, Li and colleagues demonstrate that high-performing LLMs not only predict neural responses more accurately than other LLMs but also align more closely with the hierarchical language processing pathway in the brain, revealing parallels between these models and human cognitive mechanisms.

Quantum error mitigation improves the accuracy of quantum computers at a computational overhead. Liao et al. demonstrate that classical machine learning models can deliver accuracy comparable to that of conventional techniques while reducing quantum computational costs.

Sampling rare events is key to various fields of science, but current methods are inefficient. Asghar and colleagues propose a rare event sampler based on normalizing flow neural networks that requires no prior data or collective variables, works at and out of equilibrium and keeps efficiency constant as events become rarer.

A generative model guided by a machine-learning-based classifier capable of assessing unexplored regions of the peptide space in the search for new self-assembling sequences.

Electronic skin with decoupled force feedback is essential in robotics. Yan et al. develop a soft magnetic skin capable of self-decoupling three-axis forces per taxel, reducing calibration complexity from quadratic or cubic scales to a linear scale.

A generative model that leverages a graph transformer and protein language model to generate residue sequences and full-atom structures of protein pockets is introduced, which outperforms state-of-the-art approaches.

Many physical systems involve long-range interactions, which present a considerable obstacle to large-scale simulations. Cai, Li and Wang introduce NeuralMAG, a deep learning approach to reduce complexity and accelerate micromagnetic simulations.