ProtAgents: protein discovery via large language model multi-agent collaborations combining physics and machine learning.

Designing de novo proteins beyond those found in nature holds significant promise for advancements in both scientific and engineering applications. Current methodologies for protein design often rely on AI-based models, such as surrogate models that address end-to-end problems by linking protein structure to material properties or vice versa. However, these models frequently focus on specific material objectives or structural properties, limiting their flexibility when incorporating out-of-domain knowledge into the design process or comprehensive data analysis is required. In this study, we introduce ProtAgents, a platform for de novo protein design based on Large Language Models (LLMs), where multiple AI agents with distinct capabilities collaboratively address complex tasks within a dynamic environment. The versatility in agent development allows for expertise in diverse domains, including knowledge retrieval, protein structure analysis, physics-based simulations, and results analysis. The dynamic collaboration between agents, empowered by LLMs, provides a versatile approach to tackling protein design and analysis problems, as demonstrated through diverse examples in this study. The problems of interest encompass designing new proteins, analyzing protein structures and obtaining new first-principles data - natural vibrational frequencies - via physics simulations. The concerted effort of the system allows for powerful automated and synergistic design of de novo proteins with targeted mechanical properties. The flexibility in designing the agents, on one hand, and their capacity in autonomous collaboration through the dynamic LLM-based multi-agent environment on the other hand, unleashes great potentials of LLMs in addressing multi-objective materials problems and opens up new avenues for autonomous materials discovery and design.

Multicell-Fold: geometric learning in folding multicellular life.

During developmental processes such as embryogenesis, how a group of cells fold into specific structures, is a central question in biology. However, it remains a major challenge to understand and predict the behavior of every cell within the living tissue over time during such intricate processes. Here we present a geometric deep-learning model that can accurately capture the highly convoluted interactions among cells. We demonstrate that multicellular data can be represented with both granular and foam-like physical pictures through a unified graph data structure, considering both cellular interactions and cell junction networks. Using this model, we achieve interpretable 4-D morphological sequence alignment, and predicting cell rearrangements before they occur at single-cell resolution. Furthermore, using neural activation map and ablation studies, we demonstrate cell geometries and cell junction networks together regulate morphogenesis at single-cell precision. This approach offers a pathway toward a unified dynamic atlas for a variety of developmental processes.

Generative Retrieval-Augmented Ontologic Graph and Multiagent Strategies for Interpretive Large Language Model-Based Materials Design.

Transformer neural networks show promising capabilities, in particular for uses in materials analysis, design, and manufacturing, including their capacity to work effectively with human language, symbols, code, and numerical data. Here, we explore the use of large language models (LLMs) as a tool that can support engineering analysis of materials, applied to retrieving key information about subject areas, developing research hypotheses, discovery of mechanistic relationships across disparate areas of knowledge, and writing and executing simulation codes for active knowledge generation based on physical ground truths. Moreover, when used as sets of AI agents with specific features, capabilities, and instructions, LLMs can provide powerful problem-solution strategies for applications in analysis and design problems. Our experiments focus on using a fine-tuned model, MechGPT, developed based on training data in the mechanics of materials domain. We first affirm how fine-tuning endows LLMs with a reasonable understanding of subject area knowledge. However, when queried outside the context of learned matter, LLMs can have difficulty recalling correct information and may hallucinate. We show how this can be addressed using retrieval-augmented Ontological Knowledge Graph strategies. The graph-based strategy helps us not only to discern how the model understands what concepts are important but also how they are related, which significantly improves generative performance and also naturally allows for injection of new and augmented data sources into generative AI algorithms. We find that the additional feature of relatedness provides advantages over regular retrieval augmentation approaches and not only improves LLM performance but also provides mechanistic insights for exploration of a material design process. Illustrated for a use case of relating distinct areas of knowledge, here, music and proteins, such strategies can also provide an interpretable graph structure with rich information at the node, edge, and subgraph level that provides specific insights into mechanisms and relationships. We discuss other approaches to improve generative qualities, including nonlinear sampling strategies and agent-based modeling that offer enhancements over single-shot generations, whereby LLMs are used to both generate content and assess content against an objective target. Examples provided include complex question answering, code generation, and execution in the context of automated force-field development from actively learned density functional theory (DFT) modeling and data analysis.

Heterogeneous and Cooperative Rupture of Histidine-Ni2+ Metal-Coordination Bonds on Rationally Designed Protein Templates.

Metal-coordination bonds, a highly tunable class of dynamic noncovalent interactions, are pivotal to the function of a variety of protein-based natural materials and have emerged as binding motifs to produce strong, tough, and self-healing bioinspired materials. While natural proteins use clusters of metal-coordination bonds, synthetic materials frequently employ individual bonds, resulting in mechanically weak materials. To overcome this current limitation, we rationally designed a series of elastin-like polypeptide templates with the capability of forming an increasing number of intermolecular histidine-Ni2+ metal-coordination bonds. Using single-molecule force spectroscopy and steered molecular dynamics simulations, we show that templates with three histidine residues exhibit heterogeneous rupture pathways, including the simultaneous rupture of at least two bonds with more-than-additive rupture forces. The methodology and insights developed improve our understanding of the molecular interactions that stabilize metal-coordinated proteins and provide a general route for the design of new strong, metal-coordinated materials with a broad spectrum of dissipative time scales.

Crosslinker energy landscape effects on dynamic mechanical properties of ideal polymer hydrogels.

Reversible crosslinkers can enable several desirable mechanical properties, such as improved toughness and self-healing, when incorporated in polymer networks for bioengineering and structural applications. In this work, we performed coarse-grained molecular dynamics to investigate the effect of the energy landscape of reversible crosslinkers on the dynamic mechanical properties of crosslinked polymer network hydrogels. We report that, for an ideal network, the energy potential of the crosslinker interaction drives the viscosity of the network, where a stronger potential results in a higher viscosity. Additional topographical analyses reveal a mechanistic understanding of the structural rearrangement of the network as it deforms and indicate that as the number of defects increases in the network, the viscosity of the network increases. As an important validation for the relationship between the energy landscape of a crosslinker chemistry and the resulting dynamic mechanical properties of a crosslinked ideal network hydrogel, this work enhances our understanding of deformation mechanisms in polymer networks that cannot easily be revealed by experiment and reveals design ideas that can lead to better performance of the polymer network at the macroscale.

Learning Dynamics from Multicellular Graphs with Deep Neural Networks.

Multicellular self-assembly into functional structures is a dynamic process that is critical in the development and diseases, including embryo development, organ formation, tumor invasion, and others. Being able to infer collective cell migratory dynamics from their static configuration is valuable for both understanding and predicting these complex processes. However, the identification of structural features that can indicate multicellular motion has been difficult, and existing metrics largely rely on physical instincts. Here we show that using a graph neural network (GNN), the motion of multicellular collectives can be inferred from a static snapshot of cell positions, in both experimental and synthetic datasets.

Robust myco-composites: a biocomposite platform for versatile hybrid-living materials.

Fungal mycelium, a living network of filamentous threads, thrives on lignocellulosic waste and exhibits rapid growth, hydrophobicity, and intrinsic regeneration, offering a potential means to create next-generation sustainable and functional composites. However, existing hybrid-living mycelium composites (myco-composites) are tremendously constrained by conventional mold-based manufacturing processes, which are only compatible with simple geometries and coarse biomass substrates that enable gas exchange. Here we introduce a class of structural myco-composites manufactured with a novel platform that harnesses high-resolution biocomposite additive manufacturing and robust mycelium colonization with indirect inoculation. We leverage principles of hierarchical composite design and selective nutritional provision to create a robust myco-composite that is scalable, tunable, and compatible with complex geometries. To illustrate the versatility of this platform, we characterize the impact of mycelium colonization on mechanical and surface properties of the composite. We found that our method yields the strongest mycelium composite reported to date with a modulus of 160 MPa and tensile strength of 0.72 MPa, which represents over a 15-fold improvement over typical mycelium composites, and further demonstrate unique applications with fabrication of foldable bio-welded containers and flexible mycelium textiles. This study bridges the gap between biocomposite and hybrid-living materials research, opening the door to advanced structural mycelium applications and demonstrating a novel platform for development of diverse hybrid-living materials.

ForceGen: End-to-end de novo protein generation based on nonlinear mechanical unfolding responses using a language diffusion model.

Through evolution, nature has presented a set of remarkable protein materials, including elastins, silks, keratins and collagens with superior mechanical performances that play crucial roles in mechanobiology. However, going beyond natural designs to discover proteins that meet specified mechanical properties remains challenging. Here, we report a generative model that predicts protein designs to meet complex nonlinear mechanical property-design objectives. Our model leverages deep knowledge on protein sequences from a pretrained protein language model and maps mechanical unfolding responses to create proteins. Via full-atom molecular simulations for direct validation, we demonstrate that the designed proteins are de novo, and fulfill the targeted mechanical properties, including unfolding energy and mechanical strength, as well as the detailed unfolding force-separation curves. Our model offers rapid pathways to explore the enormous mechanobiological protein sequence space unconstrained by biological synthesis, using mechanical features as the target to enable the discovery of protein materials with superior mechanical properties.

BioinspiredLLM: Conversational Large Language Model for the Mechanics of Biological and Bio-Inspired Materials.

The study of biological materials and bio-inspired materials science is well established; however, surprisingly little knowledge is systematically translated to engineering solutions. To accelerate discovery and guide insights, an open-source autoregressive transformer large language model (LLM), BioinspiredLLM, is reported. The model is finetuned with a corpus of over a thousand peer-reviewed articles in the field of structural biological and bio-inspired materials and can be prompted to recall information, assist with research tasks, and function as an engine for creativity. The model has proven that it is able to accurately recall information about biological materials and is further strengthened with enhanced reasoning ability, as well as with Retrieval-Augmented Generation (RAG) to incorporate new data during generation that can also help to traceback sources, update the knowledge base, and connect knowledge domains. BioinspiredLLM also has shown to develop sound hypotheses regarding biological materials design and remarkably so for materials that have never been explicitly studied before. Lastly, the model shows impressive promise in collaborating with other generative artificial intelligence models in a workflow that can reshape the traditional materials design process. This collaborative generative artificial intelligence method can stimulate and enhance bio-inspired materials design workflows. Biological materials are at a critical intersection of multiple scientific fields and models like BioinspiredLLM help to connect knowledge domains.

ForceGen: End-to-end de novo protein generation based on nonlinear mechanical unfolding responses using a protein language diffusion model.

Through evolution, nature has presented a set of remarkable protein materials, including elastins, silks, keratins and collagens with superior mechanical performances that play crucial roles in mechanobiology. However, going beyond natural designs to discover proteins that meet specified mechanical properties remains challenging. Here we report a generative model that predicts protein designs to meet complex nonlinear mechanical property-design objectives. Our model leverages deep knowledge on protein sequences from a pre-trained protein language model and maps mechanical unfolding responses to create novel proteins. Via full-atom molecular simulations for direct validation, we demonstrate that the designed proteins are novel, and fulfill the targeted mechanical properties, including unfolding energy and mechanical strength, as well as the detailed unfolding force-separation curves. Our model offers rapid pathways to explore the enormous mechanobiological protein sequence space unconstrained by biological synthesis, using mechanical features as target to enable the discovery of protein materials with superior mechanical properties.

Untapped Potential of Deep Eutectic Solvents for the Synthesis of Bioinspired Inorganic-Organic Materials.

Since the discovery of deep eutectic solvents (DESs) in 2003, significant progress has been made in the field, specifically advancing aspects of their preparation and physicochemical characterization. Their low-cost and unique tailored properties are reasons for their growing importance as a sustainable medium for the resource-efficient processing and synthesis of advanced materials. In this paper, the significance of these designer solvents and their beneficial features, in particular with respect to biomimetic materials chemistry, is discussed. Finally, this article explores the unrealized potential and advantageous aspects of DESs, focusing on the development of biomineralization-inspired hybrid materials. It is anticipated that this article can stimulate new concepts and advances providing a reference for breaking down the multidisciplinary borders in the field of bioinspired materials chemistry, especially at the nexus of computation and experiment, and to develop a rigorous materials-by-design paradigm.

Generative design of de novo proteins based on secondary structure constraints using an attention-based diffusion model.

We report two generative deep learning models that predict amino acid sequences and 3D protein structures based on secondary structure design objectives via either overall content or per-residue structure. Both models are robust regarding imperfect inputs and offer de novo design capacity as they can discover new protein sequences not yet discovered from natural mechanisms or systems. The residue-level secondary structure design model generally yields higher accuracy and more diverse sequences. These findings suggest unexplored opportunities for protein designs and functional outcomes within the vast amino acid sequences beyond known proteins. Our models, based on an attention-based diffusion model and trained on a dataset extracted from experimentally known 3D protein structures, offer numerous downstream applications in conditional generative design of various biological or engineering systems. Future work may include additional conditioning, and an exploration of other functional properties of the generated proteins for various properties beyond structural objectives.

Modeling and design of heterogeneous hierarchical bioinspired spider web structures using deep learning and additive manufacturing.

Spider webs are incredible biological structures, comprising thin but strong silk filament and arranged into complex hierarchical architectures with striking mechanical properties (e.g., lightweight but high strength, achieving diverse mechanical responses). While simple 2D orb webs can easily be mimicked, the modeling and synthesis of 3D-based web structures remain challenging, partly due to the rich set of design features. Here, we provide a detailed analysis of the heterogeneous graph structures of spider webs and use deep learning as a way to model and then synthesize artificial, bioinspired 3D web structures. The generative models are conditioned based on key geometric parameters (including average edge length, number of nodes, average node degree, and others). To identify graph construction principles, we use inductive representation sampling of large experimentally determined spider web graphs, to yield a dataset that is used to train three conditional generative models: 1) an analog diffusion model inspired by nonequilibrium thermodynamics, with sparse neighbor representation; 2) a discrete diffusion model with full neighbor representation; and 3) an autoregressive transformer architecture with full neighbor representation. All three models are scalable, produce complex, de novo bioinspired spider web mimics, and successfully construct graphs that meet the design objectives. We further propose an algorithm that assembles web samples produced by the generative models into larger-scale structures based on a series of geometric design targets, including helical and parametric shapes, mimicking, and extending natural design principles toward integration with diverging engineering objectives. Several webs are manufactured using 3D printing and tested to assess mechanical properties.

Coordination Stoichiometry Effects on the Binding Hierarchy of Histamine and Imidazole-M2+ Complexes.

Histidine-M2+ coordination bonds are a recognized bond motif in biogenic materials with high hardness and extensibility, which has led to growing interest in their use in soft materials for mechanical function. However, the effect of different metal ions on the stability of the coordination complex remains poorly understood, complicating their implementation in metal-coordinated polymer materials. Herein, rheology experiments and density functional theory calculations are used to characterize the stability of coordination complexes and establish the binding hierarchy of histamine and imidazole with Ni2+ , Cu2+ , and Zn2+ . It is found that the binding hierarchy is driven by the specific affinity of the metal ions to different coordination states, which can be macroscopically tuned by changing the metal-to-ligand stoichiometry of the metal-coordinated network. These findings facilitate the rational selection of metal ions for optimizing the mechanical properties of metal-coordinated materials.

Application of the Interagency and Modeling Analysis Group Model Verification Approach for Scientific Reproducibility in a Study of Biomineralization.

Model verification is a critical aspect of scientific accountability, transparency, and learning. Here, we demonstrate an application of a model verification approach for a molecular dynamics (MD) simulation, where the interactions between silica and silk protein were studied experimentally toward understanding biomineralization. Following the ten rules for credible modeling and simulation of biosciences as developed in Erdemir et al., the authors of the original paper collaborated with an external modeling group to verify the key findings of their original simulation model and to document this verification approach. The process resulted in successful replication of the key findings of the original model. Beyond verification, study of the model from a new perspective generated new insight into the basic assumptions. We discuss key learnings for how model validation processes can be improved more generally, specifically through improved documentation methods. We anticipate that this application of our protocol for model verification can be further replicated and improved to verify and validate other simulations.

Localization of Zn2+ ions affects the structural folding and mechanics of Nereis virens Nvjp-1.

Several biological organisms utilize metal-coordination bonds to produce remarkable materials, such as the jaw of the marine worm Nereis virens, where metal-coordination bonds yield remarkable hardness without mineralization. Though the structure of a major component of the jaw, the Nvjp-1 protein, has recently been resolved, a detailed nanostructural understanding of the role of metal ions on the structural and mechanical properties of the protein is missing, especially with respect to the localization of metal ions. In this work, atomistic replica exchange molecular dynamics with explicit water and Zn2+ ions and steered molecular dynamics simulations were used to explore how the initial localization of the Zn2+ ions impacts the structural folding and mechanical properties of Nvjp-1. We found that the initial distribution of metal ions for Nvjp-1, and likely for other proteins with high amounts of metal-coordination, has important effects on the resulting structure, with larger metal ion quantity resulting in a more compact structure. These structural compactness trends, however, are independent from the mechanical tensile strength of the protein, which increases with greater hydrogen bond content and uniform distribution of metal ions. Our results indicate that different physical principles underlie the structure or mechanics of Nvjp-1, with broader implications in the development optimized hardened bioinspired materials and the modeling of proteins with significant metal ion content.

Bond clusters control rupture force limit in shear loaded histidine-Ni2+ metal-coordinated proteins.

Dynamic noncovalent interactions are pivotal to the structure and function of biological proteins and have been used in bioinspired materials for similar roles. Metal-coordination bonds, in particular, are especially tunable and enable control over static and dynamic properties when incorporated into synthetic materials. Despite growing efforts to engineer metal-coordination bonds to produce strong, tough, and self-healing materials, the systematic characterization of the exact contribution of these bonds towards mechanical strength and the effect of geometric arrangements is missing, limiting the full design potential of these bonds. In this work, we engineer the cooperative rupture of metal-coordination bonds to increase the rupture strength of metal-coordinated peptide dimers. Utilizing all-atom steered molecular dynamics simulations on idealized bidentate histidine-Ni2+ coordinated peptides, we show that histidine-Ni2+ bonds can rupture cooperatively in groups of two to three bonds. We find that there is a strength limit, where adding additional coordination bonds does not contribute to the additional increase in the protein rupture strength, likely due to the highly heterogeneous rupture behavior exhibited by the coordination bonds. Further, we show that this coordination bond limit is also found natural metal-coordinated biological proteins. Using these insights, we quantitatively suggest how other proteins can be rationally designed with dynamic noncovalent interactions to exhibit cooperative bond breaking behavior. Altogether, this work provides a quantitative analysis of the cooperativity and intrinsic strength limit for metal-coordination bonds with the aim of advancing clear guiding molecular principles for the mechanical design of metal-coordinated materials.

3D Printability of Silk/Hydroxyapatite Composites for Microprosthetic Applications.

Micro-prosthetics requires the fabrication of mechanically robust and personalized components with sub-millimetric feature accuracy. Three-dimensional (3D) printing technologies have had a major impact on manufacturing such miniaturized devices for biomedical applications; however, biocompatibility requirements greatly constrain the choice of usable materials. Hydroxyapatite (HA) and its composites have been widely employed to fabricate bone-like structures, especially at the macroscale. In this work, we investigate the rheology, printability, and prosthetic mechanical properties of HA and HA-silk protein composites, focusing on the roles of composition and water content. We correlate key linear and nonlinear shear rheological parameters to geometric outcomes of printing and explain how silk compensates for the inherent brittleness of printed HA components. By increasing ink ductility, the inclusion of silk improves the quality of printed items through two mechanisms: (1) reducing underextrusion by lowering the required elastic modulus and, (2) reducing slumping by increasing the ink yield stress proportional to the modulus. We demonstrate that the elastic modulus and compressive strength of parts fabricated from silk-HA inks are higher than those for rheologically comparable pure-HA inks. We construct a printing map to guide the manufacturing of HA-based inks with excellent final properties, especially for use in biomedical applications for which sub-millimetric features are required.

A Molecular-Scale Understanding of Misorientation Toughening in Corals and Seashells.

Biominerals are organic-mineral composites formed by living organisms. They are the hardest and toughest tissues in those organisms, are often polycrystalline, and their mesostructure (which includes nano- and microscale crystallite size, shape, arrangement, and orientation) can vary dramatically. Marine biominerals may be aragonite, vaterite, or calcite, all calcium carbonate (CaCO3 ) polymorphs, differing in crystal structure. Unexpectedly, diverse CaCO3 biominerals such as coral skeletons and nacre share a similar characteristic: Adjacent crystals are slightly misoriented. This observation is documented quantitatively at the micro- and nanoscales, using polarization-dependent imaging contrast mapping (PIC mapping), and the slight misorientations are consistently between 1° and 40°. Nanoindentation shows that both polycrystalline biominerals and abiotic synthetic spherulites are tougher than single-crystalline geologic aragonite. Molecular dynamics (MD) simulations of bicrystals at the molecular scale reveal that aragonite, vaterite, and calcite exhibit toughness maxima when the bicrystals are misoriented by 10°, 20°, and 30°, respectively, demonstrating that slight misorientation alone can increase fracture toughness. Slight-misorientation-toughening can be harnessed for synthesis of bioinspired materials that only require one material, are not limited to specific top-down architecture, and are easily achieved by self-assembly of organic molecules (e.g., aspirin, chocolate), polymers, metals, and ceramics well beyond biominerals.

Unsupervised cross-domain translation via deep learning and adversarial attention neural networks and application to music-inspired protein designs.

Taking inspiration from nature about how to design materials has been a fruitful approach, used by humans for millennia. In this paper we report a method that allows us to discover how patterns in disparate domains can be reversibly related using a computationally rigorous approach, the AttentionCrossTranslation model. The algorithm discovers cycle- and self-consistent relationships and offers a bidirectional translation of information across disparate knowledge domains. The approach is validated with a set of known translation problems, and then used to discover a mapping between musical data-based on the corpus of note sequences in J.S. Bach's Goldberg Variations created in 1741-and protein sequence data-information sampled more recently. Using protein folding algorithms, 3D structures of the predicted protein sequences are generated, and their stability is validated using explicit solvent molecular dynamics. Musical scores generated from protein sequences are sonified and rendered into audible sound.