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 cell immunogenicity scoring via cross-domain aided predictive engine (T-SCAPE), a novel multidomain deep learning framework that leverages adversarial domain adaptation to integrate diverse immunologically relevant data sources, including major histocompatibility complex (MHC) presentation, peptide-MHC (pMHC) binding affinity, T cell receptor-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. It also accurately predicts the antidrug antibody-inducing potential of therapeutic antibodies without requiring MHC inputs. This success is attributed to T-SCAPE's biologically grounded and data-driven multidomain pretraining. Its consistent and robust performance highlights its potential to advance the development of safer and more effective vaccines and protein therapeutics.

While deep learning (DL) has brought a revolution in the protein structure prediction field, still an important question remains how the revolution can be transferred to advances in structure-based drug discovery. Because the lessons from the recent GPCR dock challenge were inconclusive primarily due to the size of the dataset, in this work we further elaborated on 70 diverse GPCR complexes bound to either small molecules or peptides to investigate the best-practice modeling and docking strategies for GPCR drug discovery. From our quantitative analysis, it is shown that substantial improvements in docking and virtual screening have been possible by the advance in DL-based protein structure predictions with respect to the expected results from the combination of best pre-DL tools. The success rate of docking on DL-based model structures approaches that of cross-docking on experimental structures, showing over 30% improvement from the best pre-DL protocols. This amount of performance could be achieved only when two modeling points were considered properly: 1) correct functional-state modeling of receptors and 2) receptor-flexible docking. Best-practice modeling strategies and the model confidence estimation metric suggested in this work may serve as a guideline for future computer-aided GPCR drug discovery scenarios.

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 cell immunogenicity scoring via cross-domain aided predictive engine (T-SCAPE), a novel multidomain deep learning framework that leverages adversarial domain adaptation to integrate diverse immunologically relevant data sources, including major histocompatibility complex (MHC) presentation, peptide-MHC (pMHC) binding affinity, T cell receptor-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. It also accurately predicts the antidrug antibody-inducing potential of therapeutic antibodies without requiring MHC inputs. This success is attributed to T-SCAPE's biologically grounded and data-driven multidomain pretraining. Its consistent and robust performance highlights its potential to advance the development of safer and more effective vaccines and protein therapeutics.

While deep learning (DL) has brought a revolution in the protein structure prediction field, still an important question remains how the revolution can be transferred to advances in structure-based drug discovery. Because the lessons from the recent GPCR dock challenge were inconclusive primarily due to the size of the dataset, in this work we further elaborated on 70 diverse GPCR complexes bound to either small molecules or peptides to investigate the best-practice modeling and docking strategies for GPCR drug discovery. From our quantitative analysis, it is shown that substantial improvements in docking and virtual screening have been possible by the advance in DL-based protein structure predictions with respect to the expected results from the combination of best pre-DL tools. The success rate of docking on DL-based model structures approaches that of cross-docking on experimental structures, showing over 30% improvement from the best pre-DL protocols. This amount of performance could be achieved only when two modeling points were considered properly: 1) correct functional-state modeling of receptors and 2) receptor-flexible docking. Best-practice modeling strategies and the model confidence estimation metric suggested in this work may serve as a guideline for future computer-aided GPCR drug discovery scenarios.

Atomic-level knowledge of protein-ligand interactions allows a detailed understanding of protein functions and provides critical clues to discovering molecules regulating the functions. While recent innovative deep learning methods for protein structure prediction dramatically increased the structural coverage of the human proteome, molecular interactions remain largely unknown. A new database, HProteome-BSite, provides predictions of binding sites and ligands in the enlarged 3D human proteome. The model structures for human proteins from the AlphaFold Protein Structure Database were processed to structural domains of high confidence to maximize the coverage and reliability of interaction prediction. For ligand binding site prediction, an updated version of a template-based method GalaxySite was used. A high-level performance of the updated GalaxySite was confirmed. HProteome-BSite covers 80.74% of the UniProt entries in the AlphaFold human 3D proteome. Predicted binding sites and binding poses of potential ligands are provided for effective applications to further functional studies and drug discovery. The HProteome-BSite database is available at https://galaxy.seoklab.org/hproteome-bsite/database and is free and open to all users.

. Both networks incorporate physical knowledge as an 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 that the physics-based components and the consensus ranking approach are the two most crucial factors contributing to its ranking success. We expect that DENOISer may assist future drug discovery endeavors by providing more accurate structural models for protein-ligand complexes.

Abstract The precise de novo design of antibodies remains a therapeutic challenge. The AI platform, GaluxDesign, validated its capabilities through two strategies: prior work showed comprehensive exploration, and this study validates a high-efficiency, precision approach. GaluxDesign was applied to eight distinct epitopes across six therapeutic targets, synthesizing and testing a focused set of 50 de novo IgG candidates per epitope. This precision-scale campaign yielded a 10.5% binder rate (estimated EC 50 < 100 nM), identifying target-specific binders for seven of eight epitopes. These novel antibodies exhibit therapeutically relevant properties, with sub-nanomolar to single-digit nanomolar dissociation constants (K d ) confirmed for multiple candidates. These findings confirm that a high-efficiency, precision-scale workflow is a viable approach for generating novel, high-affinity therapeutic antibodies.

Abstract 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 10 6 sequences, constructed by combining 10 2 designed light chain sequences with 10 4 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 design’s 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.

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 Synthesizable Hierarchical Action-space Reinforcement learning for Pareto optimization (SHARP), 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 side chain decoration-on a diverse receptor set, SHARP consistently outperforms prior methods in producing molecules at high affinity with reasonable 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.