Volume 7 Issue 4, April 2025

AI tools are increasingly used for important decisions, but they can be uncertain about specific individuals or groups. Chakraborty et al. discuss the need for better methods to assess uncertainty in high-stakes applications such as healthcare and finance, and outline a set of main challenges to provide practical guidance for AI researchers.

A systematic review of peer-reviewed AI safety research reveals extensive work on practical and immediate concerns. The findings advocate for an inclusive approach to AI safety that embraces diverse motivations and perspectives.

A machine learning force-field framework is proposed to predict the density, viscosity and ionic conductivity of liquid electrolytes with accuracy that is higher than classical force fields.

Vascular imaging of upstream branches and obstructed flow is challenging. Here Du and colleagues present an active exploration strategy to explore and reconstruct three-dimensional vascular networks.

InstaNovo, a transformer-based model, and InstaNovo+, a multinomial diffusion model, enhance de novo peptide sequencing, enabling discovery of novel peptides, improved therapeutics sequencing coverage and detection of unreported organisms in proteomics studies

Shengchao Liu et al. present ProteinDT, a deep learning approach that can incorporate domain knowledge from textual descriptions into protein representation on a large scale.

To function in the real world, autonomous robots will have to respond to unanticipated situations. A vision-language-model-based approach is proposed to solve long-horizon robotic tasks, which can adapt to a dynamic environment.

This study presents iKraph, a large-scale biomedical knowledge graph built using an award-winning natural language processing pipeline with expert-level accuracy. Using probabilistic semantic reasoning, iKraph enables automated knowledge discovery with excellent performance.

Duan et al. introduce an optimal transport approach to generate transition states, surpassing diffusion models in precision and speed. This method can facilitate the study of chemical reactions with unknown mechanisms.

The development of artificial vision for blind people has been a long-standing endeavour. Tang et al. create a wearable multimodal visual assistance system with a human-centred design, blending software and hardware innovations to enhance usability.

Bar-Lev et al. propose a high-efficiency DNA-based storage pipeline that integrates deep neural networks, error-correcting codes and safety margins, achieving a 3,200× speed improvement and a 40% accuracy gain, paving the way for commercially viable DNA data storage.

UniPMT, a multitask learning model, is presented, which integrates three key biological relationships into a unified framework for accurate peptide–MHC–TCR binding prediction.

DYNA fine-tunes genomic foundation models with disease specificity using a Siamese network. It generalizes to rare-variant test sets and replicates results in ClinVar, advancing variant effect prediction for cardiovascular diseases and RNA splicing.