Communications Chemistry

Liquid-liquid phase separation (LLPS) of biomacromolecules has been shown to influence a large variety of biological processes via the formation of membraneless organelles. Aberrant LLPS, however, has been implicated in a number of oncological and neurodegenerative disorders. Proteins related to Alzheimer's, Parkinson's diseases and other major amyloidoses have all been shown to form these condensates. Interestingly, some amyloidogenic proteins were demonstrated to form heterotypic droplets, where both molecules coexisted within the same condensate and affected each other's aggregation. While such cross-interactions have been reported numerous times under non-LLPS conditions, the heterotypic droplet assemblies were only shown for a very limited number of amyloidogenic proteins. In this work, we examine heterotypic LLPS of prion proteins and the pro-inflammatory S100A9, whose cross-interaction under non-LLPS conditions we confirmed in a previous work. The results show that this interaction under condensation-promoting conditions enhances their aggregation, leading to the formation of a higher concentration of amyloid structures. These findings suggest that the pro-inflammatory S100A9 may have a critical role in promoting prion protein aggregation via the assembly of heterotypic condensates under high molecular crowding conditions.

Cyclic peptides can engage targets beyond the reach of small molecules, but their discovery typically relies on libraries reused across targets, diversification around a known interaction motif, or computational methods limited to a handful of non-canonical building blocks. Here, we present nCycle-Forge, a simulated annealing framework that designs cyclic peptide binders by iteratively substituting building blocks and scoring each proposal by cofolding with the target. We designed, synthesized, and screened 1,472 cyclic peptides against the active-site pocket of Thrombin and the shallow p53-binding surface of MDM2 from the target sequence alone and without fixing binding mo-tifs. The library yielded chemically diverse hits against both targets in a single round without experimental optimization, with hit rates comparable to or up to 24-fold higher than in larger experimental libraries. Purified designs reached sub- to low-micromolar potency, with the best Thrombin inhibitor at Ki 0.14 M and the best MDM2 binder at EC50 2.4 M. Surface plasmon resonance confirmed direct binding, competition with or-thosteric inhibitors at both sites, and weaker binding under conditions expected to linearize the designs. nCycle-Forge requires no building block-specific parameterization, conformer generation, or retraining, and provides a target-focused route to functional cyclic peptides beyond the canonical amino acid alphabet.

The enzymatic oxidation of 5-hydroxymethylfurfural (HMF) to 2,5-furandicarboxylic acid (FDCA) is a key step in the sustainable production of polyethylene furanoate (PEF), yet the final oxidation of 5-formyl-2-furancarboxylic acid (FFCA) to FDCA remains the principal bottleneck for industrial application. Although hydroxymethylfurfural oxidases (HMFOs) can catalyze this transformation, naturally occurring enzymes with high FFCA-oxidizing activity remain largely unexplored. Here, we present an iterative computational bioprospecting campaign to identify natural HMFOs with improved activity toward this rate-limiting reaction. Across four successive rounds, the computational workflow evolved from conservative homology-based screening to large-scale sequence mining assisted by machine learning, with each iteration redesigned according to the principal limitations identified through molecular simulations and experimental characterization. This adaptive strategy led to the discovery of several natural HMFOs with substantially improved FFCA oxidation activity, including candidates whose performance under the tested conditions approached that of the engineered benchmark 8BxHMFO. Beyond the identification of promising biocatalysts, the study demonstrates the value of continuous integration of computational prediction and experimental feedback for refining enzyme discovery campaigns, and provides practical insights into balancing catalytic prioritization, sequence diversity, developability, and computational scalability in future simulation-guided bioprospecting efforts.

Membrane permeability remains a major challenge in cyclic-peptide drug discovery, where optimization often depends on estimating changes relative to measured reference compounds. Existing predictors mainly model absolute permeability, while heterogeneity across literature sources complicates pooled training and comparative evaluation. We present CycDelta, a multimodal residue-graph model trained directly on within-source permeability differences. CycDelta combines directed message passing with pretrained Uni-Mol monomer embeddings, physicochemical descriptors and assay information in a shared encoder for reference and query peptides. To evaluate this comparative formulation under controlled conditions, we establish CycDeltaBench, comprising in-distribution, scaffold-based out-of-distribution, external and zero-shot unseen-assay tests. On the out-of-distribution task, CycDelta achieved Pearson 0.69 versus 0.46 for the strongest baseline. It also led all evaluated methods on two external data sets and achieved zero-shot Spearman 0.70 versus 0.32 for the strongest baseline on an unseen assay. Residue-level attribution relates predictions to monomer contributions, while gradient-guided screening supports prioritization of permeability-enhancing analogue hypotheses. These results position CycDelta as a transferable and interpretable framework for comparative permeability prediction and cyclic-peptide optimization.

Autonomous drug discovery via generative molecular design is critically bottlenecked by the production of chemically intractable structures. Multi-trillion-scale make-on-demand libraries guarantee synthetic feasibility, but conventional virtual screening cannot efficiently navigate these vast spaces or address multi-parameter optimization. We introduce Hyper Screening X, a generative AI framework integrating structure-based design with automated synthesis hardware. By encoding deterministic reaction logic in a generative flow network, the framework evaluates only 10 million candidates to navigate an 11-trillion-compound space, jointly optimizing physicochemical properties while ensuring compatibility with automated synthesis. Prospective validation against an SLC1A5 variant with a cryptic interface yielded a 96% synthesis success rate and a 50% functional hit rate, identifying two first-in-class lead compounds. This synthesis-aware strategy advances autonomous, closed-loop drug discovery.

Transition state (TS) generation is typically evaluated by geometric similarity, yet low root-mean-square deviation (RMSD) does not establish first-order saddle-point character or reaction-path connectivity. Here we introduce Kairos-GFM, a deterministic equivariant flow-matching model that combines a two-dimensional reaction graph with the Kabsch-aligned mean of three-dimensional reactant and product geometries. Using 881 Transition1x reactions whose reference TSs passed strict saddle-point and bidirectional intrinsic reaction coordinate validation, Kairos-GFM generates one candidate in 0.0118 s with a median RMSD of 0.14 [A]. A hierarchical physical-validation framework--spanning reaction-center geometry, raw single-point energy, DFT saddle-point optimization, harmonic frequency analysis, and IRC endpoint matching reveals that low RMSD and saddle-point convergence do not guarantee the intended reaction channel. Kairos-GFM recovers the target channel in 75.94% of reactions, outperforming diffusion-based baselines but remaining below transport-based methods. These results position Kairos-GFM as a rapid initial-geometry generator for quantum-chemical refinement and provide a transferable validation standard that distinguishes geometric similarity from physical fidelity.

Proteins function through dynamic conformational ensembles rather than a single static structure, yet accurately modeling these ensembles remains challenging. Here, we introduce TopoFlow, an evolutionarily conditioned flow-matching framework for protein conformational ensemble generation. TopoFlow uses evolutionary representations from multiple sequence alignments (MSAs) to encode global conformational heterogeneity. It combines this with structural variables from a variational autoencoder (VAE) to capture local flexibility. These complementary representations are then adaptively fused through a conditional modulation module to construct unified conditioning features that guide protein conformational ensemble generation. On the ATLAS molecular dynamics (MD) benchmark, TopoFlow improved the Jensen Shannon metrics for pairwise distances (JS-PwD), radius of gyration (JS-Rg), and time-lagged independent components (JS-TIC) by 10.73%, 5.78% and 7.75%, respectively, relative to BioEmu. Notably, this performance generalizes to experiment-based evaluation, where TopoFlow exhibits strong agreement with experimental observables, including chemical shifts (CS) and small-angle X-ray scattering (SAXS). Ablation analyses further indicated that the evolutionary representations and structural latent variables contributed complementary information to ensemble generation. Our results suggest that TopoFlow, through its integration of evolutionary and structural latent information, offers a promising strategy for generating structurally plausible and conformationally informative protein ensembles.

Directed evolution and machine-learning-assisted evolution have transformed biocatalysis by enabling rapid optimisation of catalytic performance within compatible protein scaffolds. In parallel, advances in deep-learning-driven de novo protein design have expanded accessible protein architectural space, enabling the generation of functional catalysts for increasingly challenging transformations. We herein report a synergistic design-evolution strategy to encode new-to-nature regio- and enantioselective C-H amination reactions using engineered de novo haemproteins to synthesise chiral piperidines, a privileged pharmacophore in drug discovery. By integrating motif scaffolding, computational re-design, and directed evolution, we demonstrate de novo enzyme design can enable challenging, high-energy-barrier enzyme chemistry with programmable regio- and enantioselectivity, high catalytic efficiency, excellent enzyme stability, and scaffold diversity.

Ultra-large-scale virtual screening (ULVS) increasingly relies on make-on-demand chemical spaces, but efficient access to these spaces remains challenging because their scale far exceeds what can be exhaustively enumerated and docked. Here, we present REAL-SWIT, a generative ULVS framework that learns Enamine REAL Space as a synthesizable molecular distribution and couples this generator with a target-specific scoring model to enable docking-guided exploration of REAL Space at the level of complete molecule. The generative model learned transferable features of the REAL Space distribution: approximately 96% of generated molecules were found in REAL Space, and a subset of generated molecules absent from the training release appeared in later expanded releases. In computational benchmarks, REAL-SWIT identified more molecules with favorable docking scores than representative fragment-based search strategies, including cooperative building-block combinations that fragment-level prioritization tended to miss. Experimental validation for ROCK1 yielded six biochemical inhibitors among 23 synthesized compounds, including RX-3 with an IC50 of 0.17 M. These results establish generative access to make-on-demand chemical space as a practical strategy for ULVS.

Peptide offers a powerful modality for targeting protein-protein interfaces, yet their computational design remains challenging because sequence identity, backbone geometry and side-chain packing must be optimized simultaneously within a three-dimensional pocket constrained environment. Here we present PeTFlow, a pocket-conditioned all-atom peptide co-design framework that rationalizes this constrained design problem through discrete side-chain state tokenization. In PeTFlow, each residue is represented as a chemically compatible state token, enabling unified modeling of peptide sequence, conformation and all-atom interface geometry. PeTFlow achieved superior generation quality on peptide-design benchmarks and can further be applied to therapeutically relevant peptide drug design.

Liquid-liquid phase separation (LLPS) driven by intrinsically disordered protein regions underlies the formation of membraneless organelles. The role of these structures in cell biology is an active area of research, yet their physical understanding remains less developed than that of protein assemblies with well-defined, rigid tertiary structures. The amino acid composition of disordered regions, particularly the abundance of charged residues, is known to modulate phase behavior. These factors govern accessible conformations, the assembly of compact sticker clusters, and the porosity of the resulting polymer networks. In contrast, the spatial distribution of side-chain volumes remains a comparatively understudied sequence determinant. Here, we employ a model that captures the patterning of side-chain volumes and interaction strengths without specifying chemical identity to show how sequence blockiness modulates the phase-nucleation propensity, connectivity, and compactness of polypeptide-like polymer assemblies. Using a library of polymers with identical composition but permuted residue arrangements, we show that clustering of bulky residues generates cooperative sticker domains and identify the threshold blockiness required to stabilize polymer networks. Coarse-grained simulations and mean-field theory reveal that this effect arises from an interplay among steric repulsion, solvent entropy, and higher-valency interactions, whereby blocky patterns promote the formation of multi-node percolated networks. Segregation of stickers and spacers is therefore one of the sequence parameters governing LLPS which may be useful for interpreting the evolution of disordered protein sequences, high-throughput proteome screening, and engineering of stimuli-responsive biomaterials. These rules may be exploited to predetermine condensate stability, viscoelasticity, and molecular exchange rates without altering chemical functionality.

The rapid expansion of mRNA-based medicines has driven demand for robust, high-resolution analytical methods capable of characterising critical quality attributes including sequence identity, 5[prime] capping efficiency, and poly(A) tail length and heterogeneity. Here, we present two complementary liquid chromatography-mass spectrometry based workflows for mRNA characterisation, both based on site-specific digestion with the endoribonuclease MazF from Escherichia coli. A middle-up approach, optimised via specific 5[prime]-ACA cleavage enables rapid characterisation via oligoribonucleotide mass mapping providing complete sequence coverage and simultaneous assessment of 5[prime]-capping efficiency and 3[prime]-poly(A) tail length and heterogeneity. This workflow was successfully applied to three different mRNA constructs, NLuc, eGFP and FLuc mRNAs, achieving sequence coverages of 100%, 87%, and 74%, respectively. A complementary bottom-up approach, employing less specific 5[prime]-ACX cleavage enables tandem mass spectrometry-based sequencing of shorter oligoribonucleotides and detailed characterisation via mRNA sequence mapping. Applied to SARS-CoV-2 Spike Protein mRNA, this method yielded 50% sequence coverage based on unique oligoribonucleotides with MazF alone, extended to 90% by combining MazF, partial RNase T1, and partial RNase U2 digests. Together, these workflows provide a flexible, orthogonal platform for both high-throughput quality assessment and in-depth primary sequence characterisation of mRNA vaccines and therapeutics.

Early ribonucleoprotein systems likely required mechanisms to concentrate and organize RNA. Modern ribosomes use Mg2+ and evolved Lys/Arg-rich protein extensions to stabilize their RNA backbone. Before templated synthesis, however, peptide formation likely generated heterogeneous sequence populations rather than reproducible ribosomal sequences. Here, we compare Mg2+, ribosomal peptides, and statistical peptide libraries in organizing a 136-nucleotide model of the peptidyl transferase center (sPTC). Mg2+ condenses sPTC only at a large excess of positive charge following thermal annealing, producing largely arrested libraries. Acidic conditions further promote condensation. Ribosomal peptides instead promote coacervation near charge stoichiometry and form droplets that readily fuse. Several statistical peptide libraries also coacervate with sPTC, despite comprising heterogenous mixtures rather than a single defined sequence. Increasing mean positive charge favors condensation, but Lys/Arg-containing libraries undergo liquid-liquid phase separation more readily and across broader conditions than matched libraries containing the prebiotically plausible diaminopropionic acid (Dpr) and diaminobutyric acid (Dab). Atomistic computer simulations implicate Arg as a major source of this difference, as it can form more numerous and longer-lived hydrogen bonds with RNA, while competitive partitioning experiments show preferential condensate recruitment in the order Arg > Lys > Dab > Dpr. Together, our findings show that peptide-RNA coacervation can be triggered collectively by statistical peptide ensembles, providing a plausible route to protoribosomal organization without peptide sequence-specific optimization. Cationic chemistry shapes condensate formation and material properties, suggesting that amino acid alphabet formation could have broadened the conditions supporting liquid-like protoribosomal assemblies.

Cyclic peptides can engage extended protein interfaces, but sequence changes that improve developability often disrupt the selected target-bound geometry. Existing property-guided design methods lack a principled way to preserve this geometry during local multiproperty optimization. Here we present RF-PepTune, which separates structural eligibility from property prioritization. Before any property score is queried, RF-PepTune evaluates single substitutions and all position-distinct pairs induced by admitted substitutions to build a parent-specific, pair-complete admission graph. Pareto Monte Carlo tree search then optimizes predicted permeability, solubility, non-fouling, and low-haemolysis scores exclusively within this graph, and selected sequences undergo full structural re-evaluation. In a matched three-arm comparison of 334 parents across MDM2, GABARAP, and MCL1, joint property-structure success rose from 10.2% without admission to 51.5% with single-substitution admission and 77.5% with pair-complete admission; pair-complete gating improved success by 26.0 percentage points over single-substitution gating (95% CI, 20.7-31.4). Property-positive but structure-failing endpoints fell from 88.3% to 9.3%. In a separate target-balanced set, all 90 prioritized outputs improved all four model-native scores while satisfying prespecified structural criteria. This work establishes pairwise structural gating as a general strategy for constrained multi-objective molecular optimization.

Agentic systems increasingly coordinate molecular-design tools, but it is unclear which layer of the stack limits outcomes on real projects. We developed MAGI, an open modular agent that authors objectives, launches and monitors optimization, interprets structure activity relationships, and revises its strategy accordingly. MAGI generates molecules either directly through the LLM or by delegating to REINVENT4, with scoring services interchangeable behind a common contract. We tested it across nine retrospective lead-optimization campaigns from three pharmaceutical companies, replayed under fixed temporal cutoffs. Both routes produced valid structures: LLM proposals stayed closer to local chemistry and reached comparable or higher primary activity in fewer operations, whereas REINVENT explored broader chemical space. Whether a campaign met its objective depended on the predictive models, not on the generation route: attainment followed model accuracy on the chemistry proposed, dropping once that chemistry moved outside the model applicability domain. Separately, a blinded evaluation asked whether the MAGI output could pass as expert work: chemists were not able to discriminate agentic proposals from held-out compounds, and judged the SAR reasoning broadly plausible yet incomplete. Together, these results position MAGI as a coordination layer pluggable into existing computational chemistry workflows. The ceiling on real projects, however, remains currently set by scorer applicability rather than by tool orchestration.

Fluorescence microscopy is routinely used in the quantitative analysis of Giant Unilamellar Vesicles (GUVs) because it makes GUVs visible and easy to detect. Nonetheless, it requires fluorescence-equipped microscopes and, especially, the addition of tagged molecules that may introduce undesired and unaccounted variations compared to the pristine formulation. Alternatively, phase-contrast microscopy can be used to observe GUVs without labelling, yet the automatic detection of GUVs in phase-contrast images by standard image processing techniques is difficult. To overcome this limitation, we developed a framework for the automated detection and size characterization of label-free GUVs based on YOLOv11 model, a state-of-art lightweight Convolutional Neural Network (CNN) for object detection. Trained on 164 phase-contrast microscope images containing 4666 manually annotated vesicles, the model enables robust detection under low contrast-to-noise ratio and heterogeneous illumination. The trained model is implemented in a stand-alone desktop application supporting multiple analysis in a single run, pixel-to-micrometer calibration, and automated size distribution analysis. This approach achieves high detection accuracy and competitive cross-microscope generalization using YOLOv11 nano on greyscale image. Retraining scripts compatible with standard hardware are also provided to enhance adaptability. This tool enables fast and reliable analysis of GUVs without altering their composition, supporting experimental reproducibility.

Phosphorylation is a fast, reversible layer of post-translational control that directly governs protein activity, signaling flux, and cell state. Kinases and phosphatases collectively establish these phosphorylation states, but their dysregulation drives cancer, neurodegenerative diseases, and developmental disorders. Selectively manipulating the phosphorylation of individual proteins in cells remains challenging, particularly for disordered substrates. Here, we introduce Peptide-Retargeted Phosphatases (PepRePs), a genetically encoded system that fuses a peptide binder to a serine/threonine phosphatase subunit to localize phosphatase activity to a target protein. We first establish the PepReP architecture using characterized tau-binding peptides, then apply protein language model-guided peptide design to target Rab8A, PTEN, and STAT3. Across endogenous and exogenous targets, PepRePs reduce phosphorylation, with downstream effects on tau aggregation, Rab8A-associated protein interactions, and cellular proliferation following PTEN and STAT3 targeting. Sequence-based peptide design enables targeting without structural inputs, supporting application to substrates with disordered regions that challenge conventional structure-based binder design. By coupling programmable peptide binders with modular phosphatase subunits, PepRePs enable targeted control of protein phosphorylation to probe cellular signaling and develop strategies for correcting disease-associated protein dysfunction.

Biomolecular condensates are condensed assemblies of biomolecules that form through the process of liquid phase separation. Condensates in biology typically function as membraneless organelles, providing compartmentalization in the absence of a dividing lipid membrane. The molecular make-up of different condensates includes diverse multivalent proteins and nucleic acids as the primary drivers, many of them including significant fractions of intrinsically disordered regions (IDRs). To date, the sequence to phase separation relationship of IDRs has focused largely on one protein at a time, studying single-component condensate formation, or single-component partitioning into condensates. The co-phase separation and mixing of two or more IDRs is considerably more complex as both sequences can vary widely in their self- and cross-interactions, as well as their relative abundance in solution. It is unclear that a rule which predicts how a single sequence behaves will also predict what happens when two sequences are mixed. In this study, we disentangle the influence of sequence from that of composition using a set of 18 LAF-1 RGG variants that keep the same length and amino-acid composition and change only the order of the residues. This lets us vary charge patterning and, to a lesser extent, hydropathy patterning while keeping protein composition fixed. By themselves, the sequences phase separation and single-chain compaction are controlled by their degree of charge and hydropathy patterning. Within single-component condensed phases, each sequence adopts a more extended conformational ensemble, due to a more favorable, selfsolvated environment. We find that mixing two IDRs together into a condensate causes this universal scaling behavior to break, impacted by the relative interactions of the two components and overall composition of the slab. We find two different qualitative behaviors, one characterized by cooperative co-condensation when both sequences are subcritical, and the other by scaffold-client behavior when one sequence is supercritical. The scaffold-client systems generally show a high degree of demixing, while the co-condensing systems are generally quite well-mixed in the dense phase. This is surprising because even in cases where both partners have significantly different patterning parameters, they still mix. Thus, the descriptor that predicts a sequence's behavior alone can help indicate whether it will mix with or separate from a second component, but it does not fully determine the outcome on its own.

Macromolecular release from synthetic cells often relies on membrane pores or channels. We explored whether internally generated osmotic stress could increase membrane permeability and promote release. Using double-emulsion microfluidics, we encapsulated an Escherichia coli cell-free expression system in Pluronic L121 polymersomes to co-express the fluorescent protein mScarlet3 and the dextranase PsDex1711 in the presence of dextran. PsDex1711-mediated dextran hydrolysis produced a greater increase in bulk osmolality than reactions without PsDex1711 DNA. Tracking hundreds of individual polymersomes showed that PsDex1711-containing populations expanded, whereas control populations contracted. The distribution of normalized fluorescence revealed a subset with lower residual fluorescence than the main population. In selected expanded polymersomes, integrated internal mScarlet3 fluorescence declined to 18-22% of peak levels, without visible membrane disruption, consistent with protein release. These findings support a proof of concept for using internal gene expression to promote macromolecular release through osmotic actuation without introducing dedicated membrane pores.

Ionization governs molecular behavior, yet predicting aqueous pKa accurately and tractably remains a fundamental challenge. Rigorous ensemble methods require enumerating a protonation-state space that grows exponentially with the number of ionizable sites, while fast graph predictors return numbers without thermodynamic consistency. Here we derive an exact thermodynamic identity showing that the macroscopic first-dissociation constant is fully determined by protonated-microstate populations and unique first-deprotonation events, eliminating the deprotonated ensemble algebraically and reducing aggregation to O(n+E) operations. We implement this identity in DTi-pKa, a dual-head graph neural network whose free-energy and dissociation heads are coupled by thermodynamic constraints. On four external benchmarks (355 molecules), DTi-pKa achieves a pooled MAE of 0.5665 pKa units, with 0.6285 on the 169 records lacking training-cache matches, while providing site-resolved micro-pKa values and protonation-state populations. Controlled ablations show that constraint placement at the level of the reported macroscopic equilibrium matters more than microscopic label precision--a design principle transferable to physics-informed machine learning beyond chemistry. Closure diagnostics expose numerical self-consistency as an open frontier, which we report transparently. By replacing an intractable computation with a physical identity, DTi-pKa unifies macroscopic accuracy and microscopic interpretability in a single framework.