De novo peptide design is a rapidly evolving field with significant promise for applications in both biological and biomedical sectors. Traditional peptide design methods often rely on sequence homology, which is limited to evolutionarily related proteins and fails to incorporate the key physicochemical factors necessary for proper protein folding. Generative machine learning models offer a unique opportunity to create novel peptide sequences that are inspired by existing data but are not constrained by evolutionary limitations. In this study, we developed a custom generative adversarial network (GAN) designed specifically to generate peptide sequences capable of folding into β-hairpin secondary structures. This model is built on the principles of physicochemical properties, such as hydrophobicity and residue volume, and utilizes sequence data from known structures in the PDB to guide its design process. The β-GAN model demonstrated a remarkable ability to differentiate β-hairpin structures from α-helices and disordered peptides, achieving an accuracy rate of up to 96%. Furthermore, the model produced β-hairpin peptide sequences with sequence identities as low as 31% compared to the current PDB and 50% compared to non-redundant databases. These results highlight the potential of machine learning-based generative models to push the boundaries of peptide design, creating novel sequences and structures that transcend evolutionary constraints and offer exciting new possibilities in the field.

The widespread establishment of cryo-electron microscopy (cryo-EM) facilities equipped with electron-counting direct electron detectors (DEDs) offers new opportunities for their application in microED experiments. Electron counting offers significant improvements in sensitivity necessary for detecting high-resolution diffraction spots. The limited linear range of most frame-based electron-counting DEDs makes them prone to coincidence loss from the higher intensity of low-resolution reflections. Recent advancements in event-based electron counting (EBEC) technology, as demonstrated by the Direct Electron Apollo detector, significantly reduce coincidence loss and allows it to capture stronger reflections without saturating (Figure 1). Apollo uses a novel event counting monolithic active pixel (MAPS) sensor with on-chip CDS and edge computing to count, upsample and sum the data within the camera hardware enabling much higher counting rates. Initial small-molecule microED experiments using the Apollo detector of sodium glutamate and histidine resolved to a remarkable 0.5 Å (Figure 2). These high-resolution datasets were acquired approximately ten times faster than those obtained with traditional frame-based MAPS detectors. Apollo's improved sensitivity and acquisition speed streamline experimental workflows and support high-throughput crystallographic screening, significantly expanding the practical applications of MicroED in structural biology and pharmaceutical research.

Bacillus thuringiensis (Bt) is a soil-dwelling bacteria that forms small crystalline inclusions within the confines of its cell wall as a response to starvation. These inclusions are composed exclusively of proteins with highly specific insecticidal activity, leading them to become the most widespread bio-pesticide worldwide. Unpicking the activity of these proteins requires a detailed understanding of both their structure and their arrangement within the crystal lattice. The small size of these inclusions makes them challenging for conventional structural biology as they are difficult to locate and their small number of unit cells produces much weaker Bragg scattering for a given flux. Electron microscopy can readily image nanocrystals and the larger scattering cross section of electrons relative to X-rays allows meaningful diffraction signals to be measured from far fewer crystalline repeats. In this work, we explore electron diffraction (specifically microED and serialED) as a tool for structure determination of Bt toxins in their native state. We demonstrate that electron diffraction methods are comparable and in some cases surpass other cutting-edge crystallographic methods such as serial femtosecond crystallography at Free electron lasers. The workflow presented should also apply to other sub-micron crystals that are highly beam-sensitive and difficult to study by other means.

We use both lossy and lossless compression methods to reduce large amounts of digital data, making it easier to store and manage. Lossy and lossless compression is frequently used in astronomy, to condense the large amount of data collected by instruments namely, the James Webb Space Telescope and the Hubble Space Telescope. Similarly to diffraction images, astronomical images have large areas of empty space since stellar objects are very bright, yet show very small in images, which after a while, takes up a needless amount of limited digital storage space. Lossy compression achieves higher compression ratios but sacrifices some data quality in the process, while lossless compression preserves the data’s original form, but results in less storage savings. This work, conducted at Brookhaven National Laboratory’s National Synchrotron Light Source-II at the AMX and FMX macromolecular beam lines, aimed to evaluate the effectiveness of the lossy compression method known as Wavelet Compression. Wavelet compression is a method of data compression that uses wavelet transforms to reduce the size of data while preserving its essential features. Crystals are repetitive molecular assemblies that create diffraction patterns when exposed to X-rays, while detectors capture the amplitude of diffraction pattern spots (reflections). Similar to astronomical images, diffraction images contain large areas of empty space. This unnecessary data can quickly consume limited digital storage. By studying and optimizing compression methods, we not only improve data management but also advancements in various scientific disciplines.

The effects of hydrostatic pressure on biomolecules are seemingly paradoxical. Rather than simply shrinking in size, biomolecules often reduce their effective volume using conformational changes that expose natural voids to the solvent. Due to hydrogen bond networks, bulk water itself is a very open structure that when disrupted, can collapse, effectively reducing its own volume. This is most evident in the well-known phenomenon of electrostriction, in which highly charged ions effectively compress surrounding layers of bulk water. Thus, both biomolecules and surrounding layers of water and ions can respond to pressure in potentially revealing ways. For this reason, hydrostatic pressure has recently proven a valuable means of populating physiologically important, but normally invisible conformational states (Wang J, et al. (2023) PNAS 120:e2215556120). Until now, we have had no way to tell, in the vast sea of biomolecules, which are pressure sensitive. To better identify such proteins and to help understand their structural behavior and physiological role, we introduce a novel structural technique: high-pressure limited proteolysis (HiP- LiP), a powerful new tool for detecting subtle conformational changes in response to pressure on a proteome-wide scale (Moran HM, et al. (2024) PRX Life 2:033011). By mapping residue-level changes to both experimental and predicted structures, we find that pressure-sensitive regions correlate with active sites and functionally relevant structural dynamics, particularly involving highly charged species. These maps are a rich source of new targets for structural and biophysical studies, possibly revealing unknown binding partners and associated dynamics under both normal and extreme conditions. This talk will discuss the methodology of HiP-LiP, analyze proteome-wide tends, and present multiple structural examples.

SAXS/WAXS data collection provides valuable insights into biomolecular structure and dynamics in solution, yet transforming this experimental data into accurate three-dimensional models remains challenging. We have developed a comprehensive workflow that integrates structure prediction methods, including AI prediction models such as AlphaFold, with SAXS-driven molecular dynamics (SAXS-MD) to characterize complex biomolecular assemblies. Using RNA nanoparticles (NANPs) and complex biomolecular assemblies as examples, we demonstrate how our computational pipeline reveals conformational dynamics that remain hidden in static structural techniques. The workflow combines experimental SAXS constraints with predicted structure models and refines them through SAXS-MD simulations, leveraging high-performance computing resources to achieve the necessary simulation timescales and sampling efficiency. This integrated approach identifies key dynamic regions, characterizes conformational ensembles, and elucidates structure-function relationships in systems with inherent flexibility. The synergy between experimental data, AI-driven predictions, and HPC-enabled simulations enhances our structural biology toolkit, offering complementary insights that bridge the gap between static models and dynamic biological processes.

Revealing the symmetry change across a phase transition is fundamentally important to understanding and controlling properties such as polarization (ferroelectric transition), conductivity (metal-insulator transition) and other unconventional properties including piezoelectricity, multiferr=ics, and superconductivity. The recently developed exhaustive symmetry search (ESS) technique has been proven to be an effective tool for systematically studying subtle and complex phase transitions. Here we present XsymNet, a combined machine learning and exhaustive symmetry search approach that has been developed to identify the subgroup symmetry of a material from powder diffraction. XsymNet is a convolutional neural network with multi-head attention trained on simulated diffraction data of members of the ISODISTORT subgroup tree. In this work we discuss the workflow of XsymNet, how it lowers the barrier for phase transition studies, and our future work towards automated diffraction analysis powered by ML.

Continuous Serial Electron Diffraction (SerialED) is an emerging technique with significant potential across structural biology, chemistry, pharmaceuticals, and materials science, enabling crystal structure determination from beam-sensitive samples.[1] Yet widespread adoption remains limited, in part because data processing is perceived as complex.We introduce a streamlined, Python–based processing suite tailored to continuous SerialED. Fast, robust preprocessing is followed by indexing and integration routines adapted from X-ray free electron laser (XFEL) data processing, and optimised to tackle the specific challenges of electron–diffraction data. The pipeline carries datasets seamlessly from raw diffraction frames to a refinement–ready HKL file, while comprehensive logging ensures full traceability and reproducibility.A lightweight graphical interface ties the workflow together, allowing users to launch jobs, get feedback and adjust parameters without the need for scripting. By removing technical hurdles and shortening the learning curve, the package aims to lower the barrier to routine continuous SerialED across a wide range of research fields.A beta version of the software will be provided under an MIT open-source licence.

In the quest for new drug candidates, a pivotal phase involves identifying compounds that selectively and robustly bind to their targets to modulate activity for therapeutic effects. This modulation can manifest as inhibition, activation, or allosteric regulation, among others. A core challenge in drug discovery is detecting ligands with high binding affinity to target proteins. Techniques range from high-throughput screening and computational simulations to advanced machine learning models. Fragment-based drug discovery (FBDD), particularly using X-ray crystallography beamlines, has become a prominent method for finding initial leads for small-molecule modulators. This involves soaking potential ligands into protein crystals, followed by X-ray diffraction data analysis to detect binding fragments. Despite technological advancements enhancing throughput, variations in crystals unrelated to ligand binding hinder analysis sensitivity. We present VAE-Assisted Ligand Discovery (VALDO), a novel approach to distinguish meaningful conformational changes from unrelated crystal heterogeneity. VALDO employs a variational autoencoder (VAE) to encode crystallographic reflections into a low-dimensional space, filtering out noise and reconstructing an apo state. This facilitates the creation of difference maps crucial for identifying ligand binding. Comparative benchmarks against methods like PanDDA and Cluster4x show VALDO's superior ability to detect and estimate the pose of bound drug fragments.

CRISPR-Cas12a has become a widely used tool for genome editing and molecular diagnostics. However, unintended off-target DNA cleavage by Cas12a remains a key limitation. To address this, our lab engineered a Cas12a variant by modifying the bridge helix—a structurally important element—resulting in markedly reduced off-target activity. To understand the structural basis of this enhanced specificity, we used cryo-electron microscopy (cryo-EM) to determine high-resolution structures of the variant. Initial grid preparation presented challenges in sample distribution and reproducibility. We overcame these by deriving a reproducible method for preparing graphene oxide (GO) coated Quantifoil holey carbon copper grids to achieve a sufficient area covered with the GO monolayer. These GO-coated grids significantly improved particle density within the holes while maintaining optimal ice thickness. This protocol may benefit other cryo-EM studies involving difficult samples. The resulting structures reveal novel conformational features of the bridge helix variant, offering mechanistic insights into how DNA cleavage specificity is modulated. These findings advance our understanding of Cas12a function and provide a structural framework for the rational design of next-generation genome editing tools with improved precision due to their reduced off-target effects.