Biology subjects

Hatstat, A. K.

Publications and source records attributed to Hatstat, A. K..

8 recordsLinked to original sources

bioRxiv2026

De novo design of ligand binding proteins using large language models alone

Protein design has rapidly advanced with the advent of sequence- and structure-based machine learning models. However, reasoned design, which applies physicochemical principles and rules derived from sequence-structure-function relationships, has not seen the same benefits from generative machine learning models. Here, we test the ability of common large language models (LLMs; e.g. Claude, ChatGPT, and Gemini) to consider design principles to generate de novo proteins that bind metals and lipophilic small molecules without copying existing sequences. Common LLMs alone are able to 1) generate protein sequences to adopt a desired fold and bind the target ligand and 2) explain the principles that motivate the design choices. Following structure prediction and filtering, we selected a small set of designs (6 to 12 designs per query) for experimental validation, affording metal binders in one round of LLM-based design (25% hit rate) and perfluorooctanoic acid binders in two rounds (25% hit rate in the second round of design). Importantly, the LLMs produce detailed justification to accompany the de novo designed sequences, providing a conceptual framework on which designs can be evaluated. While the successful designs have some deviations from the prompted parameters and LLM-articulated design rationale, these campaigns provide a case study that highlights the utility of LLMs in making protein design more comprehensible and accessible to users without sophisticated design expertise.

biochemistry↗

bioRxiv2026

De novo design of a protein fold for small-molecule binding through aromatic π stacking

The de novo design of proteins that bind chemically complex small molecules has broad chemical and biological implications, but strategies typically rely on a small set of protein scaffolds and require extensive experimental screening. Here, we computationally designed proteins around a minimal aromatic {pi}-stacking motif to bind the anthracycline anticancer drug doxorubicin. Experimental characterization of twelve proteins revealed a {micro}M doxorubicin binder; two additional design cycles improved scaffold stability and binding affinity to yield an 85-residue protein that binds doxorubicin with a dissociation constant of 85 nM. An X-ray crystal structure of the protein-drug complex confirmed the accuracy of the designed {pi}-{pi} stacking interactions. The designed protein could act to protect cultured cells from doxorubicin-induced cytotoxicity. Unlike previous ligand-binding protein designs based on repeat proteins or naturally occurring folds, the designed protein adopts a previously unobserved 5-helix globular fold, indicating that a broader space of folded, functional proteins exists even for compact tertiary structures smaller than 100 residues. These results demonstrate that motif-guided generative protein design can discover compact de novo protein folds capable of high-affinity recognition of chemically complex small molecules.

biochemistry↗

bioRxiv2025

BoltzGen: Toward Universal Binder Design

We introduce BoltzGen, an all-atom generative model for designing proteins and peptides across all modalities to bind a wide range of biomolecular targets. BoltzGen builds strong structural reasoning capabilities about target-binder interactions into its generative design process. This is achieved by unifying design and structure prediction, resulting in a single model that also reaches state-of-the-art folding performance. BoltzGens generation process can be controlled with a flexible design specification language over covalent bonds, structure constraints, binding sites, and more. We experimentally validate these capabilities in eight diverse design campaigns with functional and affinity readouts across 26 targets. In our experiments, binder modalities span from nanobodies to disulfide-bonded peptides, and targets from disordered proteins to small molecules. In particular, we identify nanobody binders for novel targets with low similarity to proteins with already known bound structures. We release model weights, data, and both inference and training code at: https://github.com/HannesStark/boltzgen.

bioengineering↗

bioRxiv2025

De novo design of semisynthetic protein nanopores

Protein nanopores are essential components of single-molecule oligonucleotide sequencing and sensing devices. Here, we demonstrate that installing additional de novo subunits enables large-scale architectural changes of nanopore complexes. We design de novo proteins that integrate seamlessly with the CsgG pore to form 18-subunit, 315-kilodalton complexes with precisely sculpted pore architectures and tailored ion conduction, opening new possibilities for engineering enhanced nanopores with customized structural and functional properties.

biochemistry↗

bioRxiv2024

A designed Zn2+ sensor domain transmits binding information to transmembrane histidine kinases

Generating stimulus-responsive, allosteric signaling de novo is a significant challenge in protein design. In natural systems like bacterial histidine kinases (HKs), signal transduction occurs when ligand binding initiates a signal that is amplified across biological membranes over long distances to induce large-scale rearrangements and phosphorylation relays. Here, we ask whether our understanding of protein design and multi-domain, intramolecular signaling has progressed sufficiently to enable engineering of a HK with tunable de novo components. We generated de novo metal-binding sensor domains and substituted them for the native sensor domain of a transmembrane HK, affording chimeras that transduce signals initiated from a de novo sensor. Signaling depended on the designed sensors stability and the interdomain linkers phase and length. These results show the usefulness of de novo design to elucidate biochemical mechanisms and principles for design of new signaling systems.

biochemistry↗

bioRxiv2021

Chemoproteomic-enabled characterization of small GTPase Rab1a as a target of an N-arylbenzdiimidazole ligand's rescue of Parkinson's-associated cell toxicity

The development of phenotypic models of Parkinsons disease (PD) has enabled screening and identification of phenotypically active small molecules that restore complex biological pathways affected by PD toxicity. While these phenotypic screening platforms are powerful, they do not inherently enable direct identification of the cellular targets of promising lead compounds. To overcome this, chemoproteomic platforms like Thermal Proteome Profiling (TPP) and Stability of Proteins from Rates of Oxidation (SPROX) can be implemented to reveal protein targets of biologically active small molecules. Here we utilize both of these chemoproteomic strategies to identify targets of an N-arylbenzdiimidazole compound, NAB2, which was previously identified for its ability to restore viability in cellular models of PD-associated -synuclein toxicity. The combined results from our TPP and SPROX analyses of NAB2 and the proteins in a neuroblastoma-derived SHSY5Y cell lysate reveal a previously unrecognized protein target of NAB2. This newly recognized target, Rab1a, is a small GTPase that acts as a molecular switch to regulate ER-to-Golgi trafficking, a process that is disrupted by -synuclein toxicity and restored by NAB2 treatment. Further validation reveals that NAB2 binds to Rab1a with selectivity for its GDP-bound form and that NAB2 treatment phenocopies Rab1a overexpression in alleviation of -synuclein toxicity. Finally, we conduct a preliminary investigation into the relationship between Rab1a and the E3 ubiquitin ligase, Nedd4, a previously identified NAB2 target. Together, these efforts expand our understanding of the mechanism of NAB2 in the alleviation of -synuclein toxicity and reinforce the utility of chemoproteomic identification of the targets of phenotypically active small molecules that regulate complex biological pathways.

biochemistry↗

bioRxiv2020

Bioinformatic and computational analysis reveals the prevalence and nature of PY motif-mediated protein-protein interactions in the Nedd4 family of ubiquitin ligases

The Nedd4 family contains several structurally related but functionally distinct HECT-type ubiquitin ligases. The members of the Nedd4 family are known to recognize substrates through their multiple WW domains, which recognize PY motifs (PPxY, LPxY) or phospho-threonine or phospho-serine residues. To better understand substrate specificity across the Nedd4 family, we report the development and implementation of a python-based tool, PxYFinder, to identify PY motifs in the primary sequences of previously identified interactors of Nedd4 and related ligases. Using PxYFinder, we find that, on average, half of Nedd4 family interactions are PY-motif mediated. Further, we find that PPxY motifs are more prevalent than LPxY motifs and are more likely to occur in proline-rich regions. Further, PPxY regions are more disordered on average relative to LPxY-containing regions. Informed by consensus sequences for PY motifs across the Nedd4 interactome, we rationally designed a peptide library and employed a computational screen, revealing sequence- and biomolecular interaction-dependent determinants of WW-domain/PY-motif interactions. Cumulatively, our efforts provide a new bioinformatic tool and expand our understanding of sequence and structural factors that contribute to PY-motif mediated substrate recognition across the Nedd4 family.

biochemistry↗

bioRxiv2020

Characterization of small molecule induced changes in Parkinson's-related trafficking via the Nedd4 ubiquitin signaling cascade

The benzdiimidazole NAB2 rescues -synuclein-associated trafficking defects associated with early onset Parkinsons disease in a Nedd4-dependent manner. Despite identification of E3 ubiquitin ligase Nedd4 as a putative target of NAB2, its molecular mechanism of action has not been elucidated. As such, the effect of NAB2 on Nedd4 activity and specificity was interrogated through biochemical, biophysical, and proteomic analyses. NAB2 was found to bind Nedd4 (KDapp = 42 nM), but this binding is side chain mediated and does not alter its conformation or ubiquitination kinetics in vitro. Nedd4 co-localizes with trafficking organelles, and NAB2 exposure did not alter its colocalization. Ubiquitin-enrichment coupled proteomics revealed that NAB2 stimulates ubiquitination of trafficking and transport associated proteins, most likely through modulating the substrate specificity of Nedd4, providing a putative protein network involved in the NAB2 mechanism.

biochemistry↗