Protein model quality assessment using rotation-equivariant transformations on point clouds.

Machine learning research concerning protein structure has seen a surge in popularity over the last years with promising advances for basic science and drug discovery. Working with macromolecular structure in a machine learning context requires an adequate numerical representation, and researchers have extensively studied representations such as graphs, discretized 3D grids, and distance maps. As part of CASP14, we explored a new and conceptually simple representation in a blind experiment: atoms as points in 3D, each with associated features. These features-initially just the basic element type of each atom-are updated through a series of neural network layers featuring rotation-equivariant convolutions. Starting from all atoms, we further aggregate information at the level of alpha carbons before making a prediction at the level of the entire protein structure. We find that this approach yields competitive results in protein model quality assessment despite its simplicity and despite the fact that it incorporates minimal prior information and is trained on relatively little data. Its performance and generality are particularly noteworthy in an era where highly complex, customized machine learning methods such as AlphaFold 2 have come to dominate protein structure prediction.

Protein sequence-to-structure learning: Is this the end(-to-end revolution)?

The potential of deep learning has been recognized in the protein structure prediction community for some time, and became indisputable after CASP13. In CASP14, deep learning has boosted the field to unanticipated levels reaching near-experimental accuracy. This success comes from advances transferred from other machine learning areas, as well as methods specifically designed to deal with protein sequences and structures, and their abstractions. Novel emerging approaches include (i) geometric learning, that is, learning on representations such as graphs, three-dimensional (3D) Voronoi tessellations, and point clouds; (ii) pretrained protein language models leveraging attention; (iii) equivariant architectures preserving the symmetry of 3D space; (iv) use of large meta-genome databases; (v) combinations of protein representations; and (vi) finally truly end-to-end architectures, that is, differentiable models starting from a sequence and returning a 3D structure. Here, we provide an overview and our opinion of the novel deep learning approaches developed in the last 2 years and widely used in CASP14.

Hierarchical, rotation-equivariant neural networks to select structural models of protein complexes.

Predicting the structure of multi-protein complexes is a grand challenge in biochemistry, with major implications for basic science and drug discovery. Computational structure prediction methods generally leverage predefined structural features to distinguish accurate structural models from less accurate ones. This raises the question of whether it is possible to learn characteristics of accurate models directly from atomic coordinates of protein complexes, with no prior assumptions. Here we introduce a machine learning method that learns directly from the 3D positions of all atoms to identify accurate models of protein complexes, without using any precomputed physics-inspired or statistical terms. Our neural network architecture combines multiple ingredients that together enable end-to-end learning from molecular structures containing tens of thousands of atoms: a point-based representation of atoms, equivariance with respect to rotation and translation, local convolutions, and hierarchical subsampling operations. When used in combination with previously developed scoring functions, our network substantially improves the identification of accurate structural models among a large set of possible models. Our network can also be used to predict the accuracy of a given structural model in absolute terms. The architecture we present is readily applicable to other tasks involving learning on 3D structures of large atomic systems.

Protein Connectivity in Chemotaxis Receptor Complexes.

The chemotaxis sensory system allows bacteria such as Escherichia coli to swim towards nutrients and away from repellents. The underlying pathway is remarkably sensitive in detecting chemical gradients over a wide range of ambient concentrations. Interactions among receptors, which are predominantly clustered at the cell poles, are crucial to this sensitivity. Although it has been suggested that the kinase CheA and the adapter protein CheW are integral for receptor connectivity, the exact coupling mechanism remains unclear. Here, we present a statistical-mechanics approach to model the receptor linkage mechanism itself, building on nanodisc and electron cryotomography experiments. Specifically, we investigate how the sensing behavior of mixed receptor clusters is affected by variations in the expression levels of CheA and CheW at a constant receptor density in the membrane. Our model compares favorably with dose-response curves from in vivo Förster resonance energy transfer (FRET) measurements, demonstrating that the receptor-methylation level has only minor effects on receptor cooperativity. Importantly, our model provides an explanation for the non-intuitive conclusion that the receptor cooperativity decreases with increasing levels of CheA, a core signaling protein associated with the receptors, whereas the receptor cooperativity increases with increasing levels of CheW, a key adapter protein. Finally, we propose an evolutionary advantage as explanation for the recently suggested CheW-only linker structures.

Geometric deep learning of RNA structure.

RNA molecules adopt three-dimensional structures that are critical to their function and of interest in drug discovery. Few RNA structures are known, however, and predicting them computationally has proven challenging. We introduce a machine learning approach that enables identification of accurate structural models without assumptions about their defining characteristics, despite being trained with only 18 known RNA structures. The resulting scoring function, the Atomic Rotationally Equivariant Scorer (ARES), substantially outperforms previous methods and consistently produces the best results in community-wide blind RNA structure prediction challenges. By learning effectively even from a small amount of data, our approach overcomes a major limitation of standard deep neural networks. Because it uses only atomic coordinates as inputs and incorporates no RNA-specific information, this approach is applicable to diverse problems in structural biology, chemistry, materials science, and beyond.

Molecular mechanism of biased signaling in a prototypical G protein-coupled receptor.

Biased signaling, in which different ligands that bind to the same G protein-coupled receptor preferentially trigger distinct signaling pathways, holds great promise for the design of safer and more effective drugs. Its structural mechanism remains unclear, however, hampering efforts to design drugs with desired signaling profiles. Here, we use extensive atomic-level molecular dynamics simulations to determine how arrestin bias and G protein bias arise at the angiotensin II type 1 receptor. The receptor adopts two major signaling conformations, one of which couples almost exclusively to arrestin, whereas the other also couples effectively to a G protein. A long-range allosteric network allows ligands in the extracellular binding pocket to favor either of the two intracellular conformations. Guided by this computationally determined mechanism, we designed ligands with desired signaling profiles.

A preclinical microbeam facility with a conventional x-ray tube.

PURPOSE: Microbeam radiation therapy is an innovative treatment approach in radiation therapy that uses arrays of a few tens of micrometer wide and a few hundreds of micrometer spaced planar x-ray beams as treatment fields. In preclinical studies these fields efficiently eradicated tumors while normal tissue could effectively be spared. However, development and clinical application of microbeam radiation therapy is impeded by a lack of suitable small scale sources. Until now, only large synchrotrons provide appropriate beam properties for the production of microbeams. METHODS: In this work, a conventional x-ray tube with a small focal spot and a specially designed collimator are used to produce microbeams for preclinical research. The applicability of the developed source is demonstrated in a pilot in vitro experiment. The properties of the produced radiation field are characterized by radiochromic film dosimetry. RESULTS: 50 µm wide and 400 µm spaced microbeams were produced in a 20 × 20 mm2 sized microbeam field. The peak to valley dose ratio ranged from 15.5 to 30, which is comparable to values obtained at synchrotrons. A dose rate of up to 300 mGy/s was achieved in the microbeam peaks. Analysis of DNA double strand repair and cell cycle distribution after in vitro exposures of pancreatic cancer cells (Panc1) at the x-ray tube and the European Synchrotron leads to similar results. In particular, a reduced G2 cell cycle arrest is observed in cells in the microbeam peak region. CONCLUSIONS: At its current stage, the source is restricted to in vitro applications. However, moderate modifications of the setup may soon allow in vivo research in mice and rats.

High-speed recording of neural spikes in awake mice and flies with a fluorescent voltage sensor.

Genetically encoded voltage indicators (GEVIs) are a promising technology for fluorescence readout of millisecond-scale neuronal dynamics. Previous GEVIs had insufficient signaling speed and dynamic range to resolve action potentials in live animals. We coupled fast voltage-sensing domains from a rhodopsin protein to bright fluorophores through resonance energy transfer. The resulting GEVIs are sufficiently bright and fast to report neuronal action potentials and membrane voltage dynamics in awake mice and flies, resolving fast spike trains with 0.2-millisecond timing precision at spike detection error rates orders of magnitude better than previous GEVIs. In vivo imaging revealed sensory-evoked responses, including somatic spiking, dendritic dynamics, and intracellular voltage propagation. These results empower in vivo optical studies of neuronal electrophysiology and coding and motivate further advancements in high-speed microscopy.