Mechanical force redistributes reactivity through anisotropic, topology-dependent load transmission. Predicting which covalent bond in a complex molecule ruptures first under tension remains a central problem in mechanochemistry, biomaterials and chemically recyclable polymers. Bond dissociation energy is an incomplete proxy because it describes zero-force thermodynamic stability, whereas mechanochemical site selectivity is governed more directly by the peak force sustained along a prescribed loading path. Here we introduce covalent-bond peak force network (CBPFNet), a graph-attention model that predicts covalent-bond peak force (CBPForce) from relaxed molecular structures through a virtual-stretching workflow and avoids the hundreds of constrained geometry optimisations typically required for each candidate bond. In the CHON benchmark studied here, CBPFNet reproduces bond-ranking trends from density functional theory (DFT) and can be used for bond-level screening. Post hoc attribution analyses indicate that rupture selectivity depends on both bond identity and local stiffness contrast, with relatively rigid segments concentrating tensile stress on softer adjacent bonds. In oligopeptides, the model recovers the DFT weakest bond and highlights a serine-containing motif as a possible topology-dependent weak point under the same protocol. CBPFNet therefore extends screening beyond equilibrium bond stability toward bond-resolved peak-force ranking in sustainable polymer design and biomaterials.

A research team led by UCLA and the University of Rochester has demonstrated a promising evolution of an imaging system designed to capture details within “complex media,” which scatter light, from depicting structures inside body tissue to seeing obstacles through heavy fog. The system uses physics-based machine learning to improve upon an existing imaging technique. 

In tests with standard calibration images obscured by complex media, the new system more than doubled the signal-to-noise ratio compared to a previous generation of the technology. The system also created images in close to real time — thousandths of a second.

New research found that the century-old yardstick used across many industries to estimate air-conditioning and refrigeration energy demand — called cooling degree days — is missing something fundamental. The study reports a new, physics-based version of the metric — effective cooling degree days — that captures how temperature and humidity together impact the actual work a cooling system is required to do. Applying it across North America, they found that cooling efficiency has been quietly declining by 2–4% per decade since 1971, and that the old temperature-only metric misjudges how much cooling is required in different regions, overstating demand in some places and understating it in others.

Computationally designed tellurium catalysts achieve enantioselective chalcogen-bonding catalysis using neutral, non-covalent interactions.