Volume 3 Issue 12, December 2023

The electrocatalytic nitrogen reduction reaction is a promising alternative to the Haber–Bosch process. However, the reproducibility and reliability of this process suffer from the persistence of false positives. Computational tools have the potential to alleviate this issue but several challenges must be addressed.

Machine learning interatomic potentials (MLIPs) enable materials simulations at extended length and time scales with near-ab initio accuracy. They have broad applications in the study and design of materials. Here, we discuss recent advances, challenges, and the outlook for MLIPs.

Dr Paulien Hogeweg — professor of bioinformatics at Utrecht University, who in the 1970s, together with Ben Hesper, coined the term ‘bioinformatics’ — talks to Nature Computational Science about her work on the Cellular Potts model, the integration of spatial information in modeling approaches, and her ongoing research on multilevel evolution.

EmerGNN is a flow-based graph neural network (GNN) approach that advances on conventional methodologies for predicting drug–drug interactions in emerging drugs by effectively leveraging biomedical networks.

It is difficult to identify stable surface reconstructions of complex materials. Now a Monte Carlo sampling strategy is coupled with a machine learning interatomic potential that is iteratively improved via active learning during the search.

Enzymatic pathways control a host of cellular processes, but the complexity of such pathways has made them difficult to predict. Elektrum combines neural architecture search, kinetic models and transfer learning to effectively discover CRISPR–Cas9 cleavage kinetics.

The visualization and analysis of biological events using fluorescence microscopy is limited by the noise inherent in the images obtained. Now, a self-supervised spatial redundancy denoising transformer is proposed to address this challenge.

A graph-based contrastive learning framework, LACL, is proposed for geometric domain-agnostic prediction of molecular properties to alleviate the need for molecular geometry relaxation, enabling large-scale inference scenarios.

EmerGNN, a method to predict interactions for emerging drugs, may improve patient care and drug development by providing insight into the effects of using biomedical networks in interaction predictions.

VSSR-MC is a Markov chain method based on virtual adsorption sites that interfaces with a neural network force field to provide fast, accurate and comprehensive sampling of material surfaces.

A diffusion model that generates chemical reactions in 3D with all desired symmetries preserved is established and shown to reduce transition state search from days to seconds and complement intuition-based reaction exploration with generative AI.

Developing predictive mechanistic models in biology is challenging. Elektrum uses neural architecture search, kinetic models and transfer learning to discover CRISPR–Cas9 cleavage kinetics, achieving high performance and biophysical interpretability.

SRDTrans is a self-supervised denoising method for fluorescence images powered by spatial redundancy sampling and a dedicated transformer network that achieves good performance on fast dynamics and various imaging modalities.