Demis Hassabis

This paper introduces AlphaProtein Novo, a machine-learning pipeline for de novo enzyme design that outperforms natural sequence mining in creating enzymes for new-to-nature chemistry and substrates.

AlphaGenome predicts functional genomic measurements from DNA sequences with high resolution and accuracy across multiple modalities, outperforming existing methods in variant effect prediction.

3D-Beacons provides a federated network to unify access to protein structure data from multiple sources, enabling comparative analysis of models.

AlphaFold trained on complexes, with cross-chain paired alignments and symmetry-aware losses, gets 70% of recent heteromeric interfaces acceptable or better (DockQ ≥ 0.23), 27 points above linker-joined AlphaFold.

AlphaFold, an attention network over sequence alignments and residue pairs, rivaled experimental accuracy in most CASP14 cases: median backbone r.m.s.d.95 0.96 Å versus 2.8 Å for the runner-up.

Training a network only to predict its visual input makes head direction, boundary vector, place and egocentric boundary cell responses appear, and lets agents reuse behaviours in unfamiliar environments.

The paper proposes a new theory of reward-based learning where the prefrontal cortex functions as a meta-reinforcement learning system trained by the dopamine system.

This study investigates how fast and slow escape decisions in humans are mediated by distinct fear circuits using neuroimaging and computational modeling.