Biology subjects

Seok, C.

Publications and source records attributed to Seok, C..

15 recordsLinked to original sources

bioRxiv2026

ProMiSE: Protein Multi-State Evaluation Benchmark in Biological Contexts

Proteins are inherently dynamic, with biological functions often emerging from transitions between multiple conformational states. While recent breakthroughs have largely addressed the static structure prediction problem, no systematic benchmark exists to demonstrate how well current models capture functionally relevant dynamics. We introduce ProMiSE, the first benchmark that provides both a dataset and an evaluation scheme, based on native biological assemblies and integrating major conformational change mechanisms--intrinsic, ligand-induced, and protein-induced--within a single curated dataset. We conducted a comprehensive evaluation of state-of-the-art structure prediction models, including Al-phaFold3 and recent generative approaches. Our findings reveal that current models exhibit a limited ability to sample intrinsic multi-states and are often insensitive to biological context in induced scenarios. Internal representation analysis suggests that training-data exposure can shift predictions toward dominant conformational states over alternative biologically relevant states, primarily at the structure module. In contrast, results from BioEmu indicate that reducing decoding-stage bias can substantially improve multi-state sampling without major changes to upstream pair representations.

bioinformatics↗

bioRxiv2026

G-screen: Scalable Receptor-Aware Virtual Screening through Flexible Ligand Alignment

Virtual screening has long been a central computational tool for rational ligand discovery, enabling the systematic prioritization of candidate molecules from large chemical libraries. Although docking and related approaches that explicitly account for protein-ligand interactions have been developed and refined over several decades, achieving both reliable protein-aware interaction modeling and computational scalability remains an open challenge, particularly for ultra-large chemical spaces. Ligand-based methods are fast and robust but do not explicitly incorporate protein structure, whereas docking-based approaches model protein-ligand interactions more directly at substantially higher computational cost. Here, we present G-screen, a freely available and scalable protein-aware virtual screening framework designed for cases in which an experimentally determined or predicted reference protein-ligand complex structure is available. Rather than performing computationally intensive full docking with explicit pose sampling and optimization, G-screen rapidly generates alignment-guided pose hypotheses using a flexible global alignment algorithm (G-align). The resulting aligned poses are subsequently evaluated using protein-aware pharmacophore interactions derived from the reference complex, enabling explicit atomic-level interaction analysis while retaining the scalability and robustness of ligand-based alignment methods. Benchmarking on DUD-E, LIT-PCBA, and MUV datasets demonstrates that G-screen achieves competitive discrimination and early enrichment relative to representative ligand-based and docking-based methods, while maintaining millisecond-scale per-molecule runtimes under multi-threaded execution. These results position G-screen as a practical and scalable intermediate strategy between conventional ligand-based virtual screening and computationally intensive docking workflows for efficiently filtering ultra-large chemical libraries when a reference complex structure is available. G-screen and G-align are freely available at https://github.com/seoklab/gscreen and https://github.com/seoklab/galign, respectively. Scientific ContributionWe have developed a scalable virtual screening framework for efficiently filtering ultra-large chemical libraries using a flexible global alignment algorithm combined with protein- aware pharmacophore evaluations and alignment-guided pose hypotheses. Despite explicitly capturing atomic-level interactions, the method remains highly efficient, maintaining millisecond-scale per-molecule runtimes under parallel execution. It achieves competitive discrimination and early enrichment, serving as an intermediate strategy between conventional ligand-based virtual screening and computationally intensive docking while combining the speed of ligand-based approaches with the structural context of traditional docking.

biophysics↗

bioRxiv2025

Precision at Every Scale: Efficiency in AI-Driven De Novo Antibody Design

The precise de novo design of antibodies remains a therapeutic challenge. The AI platform, GaluxDesign, was evaluated in a high-efficiency Precision-Scale Workflow by synthesizing and testing only 50 full-length IgG candidates per epitope across eight distinct epitopes from six therapeutic targets. This campaign yielded a 10.5% binder rate (estimated EC50 < 100 nM), identifying target-specific binders for seven of eight epitopes, with multiple candidates exhibiting sub-nanomolar to single-digit nanomolar dissociation constants (Kd). We further assessed the same workflow on nine shared benchmark targets selected for external comparison, where GaluxDesign identified target-specific binders for eight of nine targets, demonstrating strong target-level performance relative to previously reported de novo antibody design approaches. Together, these results establish a high-efficiency, precision-scale workflow for generating novel, high-affinity therapeutic antibodies.

bioengineering↗

bioRxiv2025

SHARP: Generating Synthesizable Molecules via Fragment-based Hierarchical Action-space Reinforcement Learning for Pareto Optimization

Designing drug-like molecules that satisfy multiple objectives--such as high binding affinity, synthesizability, and drug-likeness--poses a complex global optimization problem over an astronomically large chemical space. Existing deep learning-based molecular generative models often treat this task as distribution modeling, relying on atom-level autoregressive actions with less consideration of explicit optimization feedback. Consequently, they frequently generate invalid structures, converge to local optima, or produce synthetically infeasible candidates. Here, we introduce SHARP (Synthesizable Hierarchical Action-space Reinforcement learning for Pareto optimization), a molecular generator that addresses these limitations via a fragment-based hierarchical action space and reinforcement learning. SHARP ensures synthetic accessibility by applying action masks guided by a pretrained Synthesizability Estimation Model (SEM). The reinforcement learning (RL) policy is trained using a composite reward function integrating docking scores, pharmacophore matching, and solvent accessibility to generate functionally relevant and experimentally tractable molecules. Furthermore, across four lead optimization tasks--fragment growing, linker design, scaffold hopping, and sidechain decoration--on a diverse receptor set, SHARP consistently outperforms prior methods in producing molecules at high affinity and synthesizability. These results demonstrate that reinforcement learning with a chemically intuitive action space design can be an effective solution to the optimization challenges in AI-driven drug discovery, offering a robust framework for rational molecular design in structure-based applications.

bioinformatics↗

bioRxiv2025

TITANiAN: Robust Prediction of T-cell Epitope Immunogenicity using Adversarial Domain Adaptation Network

T-cell immunogenicity, the ability of peptide fragments to elicit T-cell responses, is a critical determinant of the safety and efficacy of protein therapeutics and vaccines. While deep learning shows promise for in silico prediction, the scarcity of comprehensive immunogenicity data is a major challenge. We present T-SCAPE, a novel multi-domain deep learning framework that leverages adversarial domain adaptation to integrate diverse immunologically relevant data sources, including MHC presentation, peptide-MHC binding affinity, TCR-pMHC interaction, source organism information, and T-cell activation. Validated through rigorous leakage-controlled benchmarks, T-SCAPE demonstrates exceptional performance in predicting T-cell activation for specific peptide-MHC pairs. Remarkably, it also accurately predicts the ADA-inducing potential of therapeutic antibodies without requiring MHC inputs. This success is attributed to T-SCAPEs biologically grounded and data-driven multi-domain pretraining. Its consistent and robust performance highlights its potential to advance the development of safer and more effective vaccines and protein therapeutics.

bioinformatics↗

bioRxiv2025

Precise, Specific, and Sensitive De Novo Antibody Design Across Multiple Cases

The precision design of antibodies, which naturally recognize diverse molecules through six variable loops, remains a critical challenge in therapeutic molecule discovery. In this study, we demonstrate that precise, sensitive, and specific antibody design can be achieved without prior antibody information across eight distinct target proteins. For each target, binders were identified from a yeast display scFv library of approximately 106 sequences, constructed by combining 102 designed light chain sequences with 104 designed heavy chain sequences. Binders with varying binding strengths were identified for all eight targets, including a case where no experimentally resolved target protein structure was available, demonstrating the highest level of precision compared to previous de novo antibody design reports. To further validate the designed antibodies, they were characterized in the IgG format for five distinct targets. The antibodies exhibited favorable developability properties, positioning them as promising hit or lead candidates. Notably, for one target in particular, the IgG-formatted antibodies exhibited affinity, activity, and developability, comparable to a commercial antibody, highlighting the sensitivity of the design method. Furthermore, binders capable of distinguishing closely related protein subtypes or mutants were identified, demonstrating that the method can achieve high molecular specificity. Cryo-EM analysis experimentally validated the designs accuracy by confirming that the key binding interactions were precisely as designed. These findings underscore the effectiveness of precision molecular design based on atomic-accuracy structure prediction. This study establishes computational antibody design as a viable approach for generating therapeutic molecules with tailored properties, with promising potential for achieving the efficacy and safety required for successful therapeutics.

bioengineering↗

bioRxiv2025

Deep learning molecular interaction motifs from Receptor structure alone

Interactions of proteins with other molecules are often mediated by a set of critical binding motifs on their surfaces. The majority of traditional binder design relied on motifs borrowed from known binder molecules, which highly restricted its applicability to novel targets or to new binding sites. In this work, we present a deep learning network MotifGen that predicts potential binder motifs directly from receptor structures without any further supporting information. MotifGen generates motif profiles at the receptor surface for 14 types of functional groups or 6 chemical interaction classes. These profiles are not only highly human interpretable, but also provide pre-trained embedding inputs for versatile few-shot binder design applications. We demonstrate MotifGens effectiveness through its application to peptide binder design and small-molecule binding site prediction, where it either surpassed existing methods or added significant value when integrated together. We expect our motif-centric binder design strategy can facilitate discovering novel binders for challenging receptor targets.

biochemistry↗

bioRxiv2024

Physics-inspired accuracy estimator for model-docked ligand complexes

Model docking, which refers to ligand docking into the protein model structures, is becoming a promising avenue in drug discovery with the advances in artificial intelligence (AI)-based protein structure prediction. However, a significant challenge remains; even when sampling was successful in model docking, typical docking score functions fail to identify correct solutions for two-thirds of them. This discrepancy between scoring and sampling majorly arises because these scoring functions poorly tolerate minor structural inaccuracies. In this work, we propose a deep neural network named DENOISer to address the scoring challenge in model-docking scenario. In the network, ligand poses are ranked by the consensus score of two independent sub-networks: the Native-likeness prediction and the Binding energy prediction networks. Both networks incorporate physical knowledge as inductive bias in order to enhance pose discrimination power while ensuring tolerance to small interfacial structural noises. Combined with Rosetta GALigandDock sampling, DENOISer outperformed existing docking tools on the PoseBusters model-docking benchmark set as well as on a broad cross-docking benchmark set. Further analyses reveal the physics-based components and the consensus ranking approach are the two most crucial factors contributing to its ranking success. We expect DENOISer may assist future drug discovery endeavors by providing more accurate structural models for protein-ligand complexes. The network is freely available at https://github.com/HParklab/DENOISer.

bioinformatics↗

bioRxiv2024

Accurate antibody loop structure prediction enables zero-shot design of target-specific antibodies

Protein loops, characterized by their versatile structures with varying sizes and shapes, can recognize a wide range of targets with high specificity and affinity. The variable loops of the antibody complementarity-determining region (CDR) are particularly crucial for immune responses and therapeutic applications due to their effective target recognition capabilities. Accurate structure prediction of these antibody loops is essential for the efficient in silico design of target-binding antibodies for therapeutic or industrial use. However, predicting antibody loop structures is challenging due to the lack of evolutionary information from related proteins. Thus, a successful ab initio structure prediction method, which operates without structural templates or related sequences, is crucial for the effective design of antibody loop-mediated interactions. This study demonstrates that highly accurate antibody loop structure prediction enables the effective zero-shot design of target-binding antibody loops. The performance of loop design has been shown to depend on the accuracy of ab initio loop structure prediction, as tested with two versions of our design model. The high affinity, diversity, novelty, and specificity of the antibody loops designed with these new methods were validated experimentally on four target proteins.

bioengineering↗

bioRxiv2024

Efficient DNA base editing via an optimized DYW-like deaminase

CRISPR-based cytosine base editors enable precise genome editing without inducing double-stranded DNA breaks, yet traditionally depend on a limited selection of deaminases from the APOBEC/AID or TadA families. Here, we introduce SsCBE, a novel CRISPR-based cytosine base editor utilizing SsdAtox, a DYW-like deaminase derived from the toxin of Pseudomonas syringae. Strategic engineering of SsdAtox has led to remarkable improvements in the base editing efficiency (by up to 8.4-fold) and specificity for SsCBE, while concurrently reducing cytotoxicity. Exhibiting exceptional versatility, SsCBE was delivered and efficiently applied using diverse delivery methods, including the engineered virus-like particles (eVLPs). Its application has enabled targeted cytosine base editing in mouse zygotes and pioneering edits in mitochondrial DNA. The advent of SsCBE marks a significant advancement in the CRISPR toolkit, providing a versatile tool for advanced research and therapeutic strategies.

molecular biology↗

bioRxiv2024

Discovery of Highly Active Kynureninases for Cancer Immunotherapy through Protein Language Model

Overcoming the immunosuppressive tumor microenvironment is a promising strategy in anticancer therapy. L-kynurenine, a strong immunosuppressive metabolite can be degraded through kynureninases. Through homology searches and protein language models, we identified and then experimentally determined the efficacy of four top-ranked kynureninases. The catalytically most active one nearly doubles turnover number over the prior best, reducing tumor weight by 3.42 times in mouse model comparisons, and thus, presenting substantial therapeutic potential.

cancer biology↗

bioRxiv2022

ECPAS/Ecm29-Mediated 26S Proteasome Disassembly Is an Adaptive Response to Glucose Starvation

The 26S proteasome consists of loosely associated 20S catalytic and 19S regulatory complexes. Approximately half of the proteasomes in eukaryotic cells exist as free 20S complexes; however, our mechanistic and physiological understanding of what determines the ratio of 26S to 20S species remains incomplete. Here, we show that glucose starvation in mammalian cells results in the uncoupling of 26S holoenzymes into intact 20S and 19S subcomplexes. Subcomplex affinity-purification and quantitative mass spectrometry revealed that Ecm29 proteasome adaptor and scaffold (ECPAS) is a crucial mediator of this structural remodeling. The loss of ECPAS abrogated 26S dissociation, leading to decreased degradation of 20S proteasome substrates such as puromycylated polypeptides and lysine-less cyclin B. In silico modeling analysis suggested that the conformational changes of ECPAS may commence the disassembly process. ECPAS was also essential for proper endoplasmic reticulum stress response and cell survival during glucose starvation. In addition, we evaluated the role of ECPAS in vivo using the mouse xenograft model and observed that glucose-deprived tumor tissues had significantly elevated 20S proteasome levels. Collectively, our results indicate that the 20S-19S disassembly mediated by ECPAS is a novel mechanism adapting global proteolysis to physiological needs and an effective cellular strategy against proteotoxic stress.

biochemistry↗

bioRxiv2021

Dynamic Interactions of Fully Glycosylated SARS-CoV-2 Spike Protein with Various Antibodies

The spread of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) presents a public health crisis, and the vaccines that can induce highly potent neutralizing antibodies are essential for ending the pandemic. The spike (S) protein on the viral envelope mediates human angiotensin-converting enzyme 2 (ACE2) binding and thus is the target of a variety of neutralizing antibodies. In this work, we built various S trimer-antibody complex structures on the basis of the fully glycosylated S protein models described in our previous work, and performed all-atom molecular dynamics simulations to get insight into the structural dynamics and interactions between S protein and antibodies. Investigation of the residues critical for S-antibody binding allows us to predict the potential influence of mutations in SARS-CoV-2 variants. Comparison of the glycan conformations between S-only and S-antibody systems reveals the roles of glycans in S-antibody binding. In addition, we explored the antibody binding modes, and the influences of antibody on the motion of S protein receptor binding domains. Overall, our analyses provide a better understanding of S-antibody interactions, and the simulation-based S-antibody interaction maps could be used to predict the influences of S mutation on S-antibody interactions, which will be useful for the development of vaccine and antibody-based therapy.

biophysics↗

bioRxiv2020

Structure, Dynamics, Receptor Binding, and Antibody Binding of Fully-glycosylated Full-length SARS-CoV-2 Spike Protein in a Viral Membrane

The spike (S) protein of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) mediates host cell entry by binding to angiotensin-converting enzyme 2 (ACE2), and is considered the major target for drug and vaccine development. We previously built fully-glycosylated full-length SARS-CoV-2 S protein models in a viral membrane including both open and closed conformations of receptor binding domain (RBD) and different templates for the stalk region. In this work, multiple s-long all-atom molecular dynamics simulations were performed to provide deeper insight into the structure and dynamics of S protein, and glycan functions. Our simulations reveal that the highly flexible stalk is composed of two independent joints and most probable S protein orientations are competent for ACE2 binding. We identify multiple glycans stabilizing the open and/or closed states of RBD, and demonstrate that the exposure of antibody epitopes can be captured by detailed antibody-glycan clash analysis instead of a commonly-used accessible surface area analysis that tends to overestimate the impact of glycan shielding and neglect possible detailed interactions between glycan and antibody. Overall, our observations offer structural and dynamic insight into SARS-CoV-2 S protein and potentialize for guiding the design of effective antiviral therapeutics.

biophysics↗

bioRxiv2020

Modeling and Simulation of a Fully-glycosylated Full-length SARS-CoV-2 Spike Protein in a Viral Membrane

This technical study describes all-atom modeling and simulation of a fully-glycosylated full-length SARS-CoV-2 spike (S) protein in a viral membrane. First, starting from PDB:6VSB and 6VXX, full-length S protein structures were modeled using template-based modeling, de-novo protein structure prediction, and loop modeling techniques in GALAXY modeling suite. Then, using the recently-determined most occupied glycoforms, 22 N-glycans and 1 O-glycan of each monomer were modeled using Glycan Reader & Modeler in CHARMM-GUI. These fully-glycosylated full-length S protein model structures were assessed and further refined against the low-resolution data in their respective experimental maps using ISOLDE. We then used CHARMM-GUI Membrane Builder to place the S proteins in a viral membrane and performed all-atom molecular dynamics simulations. All structures are available in CHARMM-GUI COVID-19 Archive (http://www.charmm-gui.org/docs/archive/covid19), so researchers can use these models to carry out innovative and novel modeling and simulation research for the prevention and treatment of COVID-19.

biophysics↗