Yanjun Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Yanjun Li has a total of 139 co-authors. Here are the top co-authors by number of shared publications:
- Chenglong Li
University of Florida
8 shared publications Chenglong Li
Department of Medicinal Chemistry, Center for Natural Products, Drug Discovery and Development, University of Florida , Gainesville, FL 32610
7 shared publications- Gustavo Seabra
University of Florida
6 shared publications - Matthew David Disney
Department of Chemistry, The Herbert Wertheim UF Scripps Institute for Biomedical Innovation and Technology, 130 Scripps Way, Jupiter, FL 33458, USA
5 shared publications - Xiaolin Li
Hangzhou Institute of Medicine, Chinese Academy of Sciences
5 shared publications - Amirhossein Taghavi
Department of Chemistry, The Herbert Wertheim UF Scripps Institute for Biomedical Innovation and Technology, 130 Scripps Way, Jupiter, FL 33458, USA
4 shared publications
Latest Publications
- Open Access
- 26 October 2026
- Published by Oxford University Press (OUP) in Nucleic Acids Research
SMARTFlexDB: a paired apo–holo RNA structure database and analysis platform for RNA conformational flexibility and small-molecule recognition
RNA structures can undergo substantial conformational changes upon small-molecule binding, yet existing structural resources provide only static RNA–ligand snapshots without connecting ligand-bound (holo) RNA structures to corresponding ligand-free (apo) conformations. SMARTFlexDB is a curated database and analysis platform that uses paired apo–holo RNA structures to characterize conformational differences and RNA–small molecule recognition. Candidate pairs were identified from the Protein Data Bank by matching RNA chains with at least 90% sequence identity and coverage, and fully conserved binding pocket nucleotides. The current release contains 434 binding pockets from 266 holo structures, linked to 186 apo structures to yield 1187 pocket-level apo–holo pairs organized into 734 database entries, spanning 150 small molecules and 23 RNA functional classes. Each entry integrates RNA sequence and secondary-structure annotations, apo and holo three-dimensional structures, RNA–ligand atomic interactions, and quantitative descriptors of global and local conformational changes. An interactive web interface enables users to explore sequence alignments and structural overlays, compare secondary structures and RNA motifs, and upload paired apo–holo structures for customized analysis. By linking ligand-bound and ligand-free RNA structures, SMARTFlexDB provides a robust foundation for investigating RNA structural flexibility and supporting RNA-targeted small-molecule drug discovery. The database and analysis platform are available at https://aidd.rc.ufl.edu/app/smartflexdb/.
- 26 October 2026
- Published by Oxford University Press (OUP) in Nucleic Acids Research
Open AccessSMARTFlexDB: a paired apo–holo RNA structure database and analysis platform for RNA conformational flexibility and small-molecule recognition
RNA structures can undergo substantial conformational changes upon small-molecule binding, yet existing structural resources provide only static RNA–ligand snapshots without connecting ligand-bound (holo) RNA structures to corresponding ligand-free (apo) conformations. SMARTFlexDB is a curated database and analysis platform that uses paired apo–holo RNA structures to characterize conformational differences and RNA–small molecule recognition. Candidate pairs were identified from the Protein Data Bank by matching RNA chains with at least 90% sequence identity and coverage, and fully conserved binding pocket nucleotides. The current release contains 434 binding pockets from 266 holo structures, linked to 186 apo structures to yield 1187 pocket-level apo–holo pairs organized into 734 database entries, spanning 150 small molecules and 23 RNA functional classes. Each entry integrates RNA sequence and secondary-structure annotations, apo and holo three-dimensional structures, RNA–ligand atomic interactions, and quantitative descriptors of global and local conformational changes. An interactive web interface enables users to explore sequence alignments and structural overlays, compare secondary structures and RNA motifs, and upload paired apo–holo structures for customized analysis. By linking ligand-bound and ligand-free RNA structures, SMARTFlexDB provides a robust foundation for investigating RNA structural flexibility and supporting RNA-targeted small-molecule drug discovery. The database and analysis platform are available at https://aidd.rc.ufl.edu/app/smartflexdb/.
- Open Access
SMARTFlexDB: a paired apo-holo RNA structure database and analysis platform for RNA conformational flexibility and small-molecule recognition
RNA structures can undergo substantial conformational changes upon small-molecule binding, yet existing structural resources provide only static RNA-ligand snapshots without connecting ligand-bound (holo) RNA structures to corresponding ligand-free (apo) conformations. SMARTFlexDB is a curated database and analysis platform that uses paired apo-holo RNA structures to characterize conformational differences and RNA–small-molecule recognition. Candidate pairs were identified from the Protein Data Bank by matching RNA chains with at least 90% sequence identity and coverage, and fully conserved binding pocket nucleotides. The current release contains 434 binding pockets from 266 holo structures, linked to 186 apo structures to yield 1,187 pocket-level apo-holo pairs organized into 734 database entries, spanning 150 small molecules and 23 RNA functional classes. Each entry integrates RNA sequence and secondary-structure annotations, apo and holo three-dimensional structures, RNA-ligand atomic interactions, and quantitative descriptors of global and local conformational changes. An interactive web interface enables users to explore sequence alignments and structural overlays, compare secondary structures and RNA motifs, and upload paired apo-holo structures for customized analysis. By linking ligand-bound and ligand-free RNA structures, SMARTFlexDB provides a robust foundation for investigating RNA structural flexibility and supporting RNA-targeted small-molecule drug discovery. The database and analysis platform are available at https://aidd.rc.ufl.edu/app/smartflexdb/ . GRAPHICAL
Open AccessSMARTFlexDB: a paired apo-holo RNA structure database and analysis platform for RNA conformational flexibility and small-molecule recognition
RNA structures can undergo substantial conformational changes upon small-molecule binding, yet existing structural resources provide only static RNA-ligand snapshots without connecting ligand-bound (holo) RNA structures to corresponding ligand-free (apo) conformations. SMARTFlexDB is a curated database and analysis platform that uses paired apo-holo RNA structures to characterize conformational differences and RNA–small-molecule recognition. Candidate pairs were identified from the Protein Data Bank by matching RNA chains with at least 90% sequence identity and coverage, and fully conserved binding pocket nucleotides. The current release contains 434 binding pockets from 266 holo structures, linked to 186 apo structures to yield 1,187 pocket-level apo-holo pairs organized into 734 database entries, spanning 150 small molecules and 23 RNA functional classes. Each entry integrates RNA sequence and secondary-structure annotations, apo and holo three-dimensional structures, RNA-ligand atomic interactions, and quantitative descriptors of global and local conformational changes. An interactive web interface enables users to explore sequence alignments and structural overlays, compare secondary structures and RNA motifs, and upload paired apo-holo structures for customized analysis. By linking ligand-bound and ligand-free RNA structures, SMARTFlexDB provides a robust foundation for investigating RNA structural flexibility and supporting RNA-targeted small-molecule drug discovery. The database and analysis platform are available at https://aidd.rc.ufl.edu/app/smartflexdb/ . GRAPHICAL
- Open Access
Pep2Mol: 3D Molecule Generation Targeting Protein-Protein Interfaces with Diffusion Models
Protein-protein interactions (PPIs) are central to biological processes. Designing small molecules that modulate dysregulated PPIs holds strong promise for targeting undruggable proteins. However, existing structure-based drug design approaches focus on well-defined small-molecule binding pockets and struggle to generalize to large, shallow, and chemically complex PPI interfaces. Here, we introduce Pep2Mol, a diffusion-based generative model for 3D molecule design that targets orthosteric PPI sites by explicitly incorporating binding peptides or proteins as structural guidance, moving beyond conventional pocket-conditioned generation. To enable model development and benchmarking, we curate a large-scale, high-quality dataset of 10,956 experimentally resolved protein complex structure pairs, each pairing an orthosteric competitive ligand with a protein binder at overlapping receptor interfaces. Pep2Mol integrates two SE(3)-equivariant graph neural networks that encode protein-ligand and protein-peptide interactions respectively, and fuses these representations via attention-based conditioning to jointly guide the diffusion trajectory. Extensive evaluations demonstrate that Pep2Mol generates chemically valid ligands with state-of-the-art binding affinities, providing a strong foundation for small-molecule inhibitor design against challenging PPI interfaces.
Open AccessPep2Mol: 3D Molecule Generation Targeting Protein-Protein Interfaces with Diffusion Models
Protein-protein interactions (PPIs) are central to biological processes. Designing small molecules that modulate dysregulated PPIs holds strong promise for targeting undruggable proteins. However, existing structure-based drug design approaches focus on well-defined small-molecule binding pockets and struggle to generalize to large, shallow, and chemically complex PPI interfaces. Here, we introduce Pep2Mol, a diffusion-based generative model for 3D molecule design that targets orthosteric PPI sites by explicitly incorporating binding peptides or proteins as structural guidance, moving beyond conventional pocket-conditioned generation. To enable model development and benchmarking, we curate a large-scale, high-quality dataset of 10,956 experimentally resolved protein complex structure pairs, each pairing an orthosteric competitive ligand with a protein binder at overlapping receptor interfaces. Pep2Mol integrates two SE(3)-equivariant graph neural networks that encode protein-ligand and protein-peptide interactions respectively, and fuses these representations via attention-based conditioning to jointly guide the diffusion trajectory. Extensive evaluations demonstrate that Pep2Mol generates chemically valid ligands with state-of-the-art binding affinities, providing a strong foundation for small-molecule inhibitor design against challenging PPI interfaces.
- Open Access
Geometric Deep Learning Reveals Ligandable and Cryptic RNA Binding Small Molecule Pockets (SMARTPocket)
RNAs are important therapeutic targets, however identifying ligandable small-molecule binding pockets remains a major barrier to RNA-targeted drug discovery. Here, SMARTPocket, an atomic-level geometric deep learning framework for predicting RNA-small molecule binding pockets directly from three-dimensional structure is introduced. SMARTPocket represents RNA as full-atom point clouds and uses transfer learning from more than 110,000 protein binding interface structures to overcome the limited number of experimentally elucidated RNA-ligand complexes. Across four established single-chain benchmarks and three broader curated benchmarks, SMARTPocket consistently outperforms existing RNA pocket predictors and general biomolecular modeling approaches. The model generalizes to apo RNA structures when conformational changes are modest, identifies cryptic ligandable pockets, and recapitulates experimentally validated binding sites in the SARS-CoV-2 frameshifting element and an RNA aptamer evolved to bind small molecules. SMARTPocket-guided docking further improves near-native RNA-ligand pose recovery and computational efficiency compared with blind docking. These results establish SMARTPocket as a generalizable framework for structure-based identification of ligandable RNA pockets and for accelerating discovery of RNA-targeted small molecules.
Open AccessGeometric Deep Learning Reveals Ligandable and Cryptic RNA Binding Small Molecule Pockets (SMARTPocket)
RNAs are important therapeutic targets, however identifying ligandable small-molecule binding pockets remains a major barrier to RNA-targeted drug discovery. Here, SMARTPocket, an atomic-level geometric deep learning framework for predicting RNA-small molecule binding pockets directly from three-dimensional structure is introduced. SMARTPocket represents RNA as full-atom point clouds and uses transfer learning from more than 110,000 protein binding interface structures to overcome the limited number of experimentally elucidated RNA-ligand complexes. Across four established single-chain benchmarks and three broader curated benchmarks, SMARTPocket consistently outperforms existing RNA pocket predictors and general biomolecular modeling approaches. The model generalizes to apo RNA structures when conformational changes are modest, identifies cryptic ligandable pockets, and recapitulates experimentally validated binding sites in the SARS-CoV-2 frameshifting element and an RNA aptamer evolved to bind small molecules. SMARTPocket-guided docking further improves near-native RNA-ligand pose recovery and computational efficiency compared with blind docking. These results establish SMARTPocket as a generalizable framework for structure-based identification of ligandable RNA pockets and for accelerating discovery of RNA-targeted small molecules.
- Open Access
- 4 May 2026
- Published by Oxford University Press (OUP) in Briefings in Bioinformatics
- Vol. 27 (3)
STORM: spatial transcriptomics optimization by resolution via matrix factorization
Classic RNA sequencing dissociates cells from their native tissue architecture, discarding spatial information that critically shapes transcriptional programs in development, homeostasis, and cancer. However, current ST platforms often produce incomplete and noisy profiles due to technical limitations and tissue variability. These limitations obscure biologically meaningful spatial patterns and hinder downstream interpretation. Here, we introduce STORM (spatial transcriptomics optimization by resolution via matrix factorization), a machine learning framework that improves the fidelity of spatial transcriptomics data under severe sparsity. STORM formulates spatial transcriptomics recovery as a low-rank tensor decomposition problem and integrates multimodal biological priors through a principled regularization strategy. Specifically, the model jointly captures spatial continuity, tissue morphology derived from whole-slide histology images, and gene–gene interaction structure informed by protein–protein interaction networks. This method enables accurate reconstruction at unobserved locations while preserving biologically meaningful spatial structure. Across diverse lung tissue profiles, including both healthy and malignant samples, STORM consistently outperforms existing state-of-the-art methods in recovering spatial gene–expression patterns and remains robust even when a majority of spatial measurements are missing. By explicitly embedding biological structure into the reconstruction process, STORM provides a reliable foundation for high-resolution spatial transcriptomic analysis in settings where experimental data are sparse or incomplete. Availability: The source code developed in this study is publicly available at https://github.com/denizgurarslan/STORM.
- 4 May 2026
- Published by Oxford University Press (OUP) in Briefings in Bioinformatics
- Vol. 27 (3)
Open AccessSTORM: spatial transcriptomics optimization by resolution via matrix factorization
Classic RNA sequencing dissociates cells from their native tissue architecture, discarding spatial information that critically shapes transcriptional programs in development, homeostasis, and cancer. However, current ST platforms often produce incomplete and noisy profiles due to technical limitations and tissue variability. These limitations obscure biologically meaningful spatial patterns and hinder downstream interpretation. Here, we introduce STORM (spatial transcriptomics optimization by resolution via matrix factorization), a machine learning framework that improves the fidelity of spatial transcriptomics data under severe sparsity. STORM formulates spatial transcriptomics recovery as a low-rank tensor decomposition problem and integrates multimodal biological priors through a principled regularization strategy. Specifically, the model jointly captures spatial continuity, tissue morphology derived from whole-slide histology images, and gene–gene interaction structure informed by protein–protein interaction networks. This method enables accurate reconstruction at unobserved locations while preserving biologically meaningful spatial structure. Across diverse lung tissue profiles, including both healthy and malignant samples, STORM consistently outperforms existing state-of-the-art methods in recovering spatial gene–expression patterns and remains robust even when a majority of spatial measurements are missing. By explicitly embedding biological structure into the reconstruction process, STORM provides a reliable foundation for high-resolution spatial transcriptomic analysis in settings where experimental data are sparse or incomplete. Availability: The source code developed in this study is publicly available at https://github.com/denizgurarslan/STORM.
- Open Access
- 14 March 2026
- Published by Association for the Advancement of Artificial Intelligence (AAAI) in Proceedings of the AAAI Conference on Artificial Intelligence
- Vol. 40 (2) , 1614-1622
Apo2Mol: 3D Molecule Generation via Dynamic Pocket-Aware Diffusion Models
Deep generative models are rapidly advancing structure-based drug design, offering substantial promise for generating small molecule ligands that bind to specific protein targets. However, most current approaches assume a rigid protein binding pocket, neglecting the intrinsic flexibility of proteins and the conformational rearrangements induced by ligand binding, limiting their applicability in practical drug discovery. Here, we propose Apo2Mol, a diffusion-based generative framework for 3D molecule design that explicitly accounts for conformational flexibility in protein binding pockets. To support this, we curate a dataset of over 24,000 experimentally resolved apo-holo structure pairs from the Protein Data Bank, enabling the characterization of protein structure changes associated with ligand binding. Apo2Mol employs a full-atom hierarchical graph-based diffusion model that simultaneously generates 3D ligand molecules and their corresponding holo pocket conformations from input apo states. Empirical studies demonstrate that Apo2Mol can achieve state-of-the-art performance in generating high-affinity ligands and accurately capture realistic protein pocket conformational changes.
- 14 March 2026
- Published by Association for the Advancement of Artificial Intelligence (AAAI) in Proceedings of the AAAI Conference on Artificial Intelligence
- Vol. 40 (2) , 1614-1622
Open AccessApo2Mol: 3D Molecule Generation via Dynamic Pocket-Aware Diffusion Models
Deep generative models are rapidly advancing structure-based drug design, offering substantial promise for generating small molecule ligands that bind to specific protein targets. However, most current approaches assume a rigid protein binding pocket, neglecting the intrinsic flexibility of proteins and the conformational rearrangements induced by ligand binding, limiting their applicability in practical drug discovery. Here, we propose Apo2Mol, a diffusion-based generative framework for 3D molecule design that explicitly accounts for conformational flexibility in protein binding pockets. To support this, we curate a dataset of over 24,000 experimentally resolved apo-holo structure pairs from the Protein Data Bank, enabling the characterization of protein structure changes associated with ligand binding. Apo2Mol employs a full-atom hierarchical graph-based diffusion model that simultaneously generates 3D ligand molecules and their corresponding holo pocket conformations from input apo states. Empirical studies demonstrate that Apo2Mol can achieve state-of-the-art performance in generating high-affinity ligands and accurately capture realistic protein pocket conformational changes.
- Open Access
Apo2Mol: 3D Molecule Generation via Dynamic Pocket-Aware Diffusion Models
Deep generative models are rapidly advancing structure-based drug design, offering substantial promise for generating small molecule ligands that bind to specific protein targets. However, most current approaches assume a rigid protein binding pocket, neglecting the intrinsic flexibility of proteins and the conformational rearrangements induced by ligand binding, limiting their applicability in practical drug discovery. Here, we propose Apo2Mol, a diffusion-based generative framework for 3D molecule design that explicitly accounts for conformational flexibility in protein binding pockets. To support this, we curate a dataset of over 24,000 experimentally resolved apo-holo structure pairs from the Protein Data Bank, enabling the characterization of protein structure changes associated with ligand binding. Apo2Mol employs a full-atom hierarchical graph-based diffusion model that simultaneously generates 3D ligand molecules and their corresponding holo pocket conformations from input apo states. Empirical studies demonstrate that Apo2Mol can achieve state-of-the-art performance in generating high-affinity ligands and accurately capture realistic protein pocket conformational changes.
Open AccessApo2Mol: 3D Molecule Generation via Dynamic Pocket-Aware Diffusion Models
Deep generative models are rapidly advancing structure-based drug design, offering substantial promise for generating small molecule ligands that bind to specific protein targets. However, most current approaches assume a rigid protein binding pocket, neglecting the intrinsic flexibility of proteins and the conformational rearrangements induced by ligand binding, limiting their applicability in practical drug discovery. Here, we propose Apo2Mol, a diffusion-based generative framework for 3D molecule design that explicitly accounts for conformational flexibility in protein binding pockets. To support this, we curate a dataset of over 24,000 experimentally resolved apo-holo structure pairs from the Protein Data Bank, enabling the characterization of protein structure changes associated with ligand binding. Apo2Mol employs a full-atom hierarchical graph-based diffusion model that simultaneously generates 3D ligand molecules and their corresponding holo pocket conformations from input apo states. Empirical studies demonstrate that Apo2Mol can achieve state-of-the-art performance in generating high-affinity ligands and accurately capture realistic protein pocket conformational changes.
- 12 November 2025
- Published by American Association for the Advancement of Science (AAAS) in Science Translational Medicine
- Vol. 17 (824) , eadn3993
Extracellular vesicle–mediated gene editing for the treatment of nonsyndromic progressive hearing loss in adult mice
The clinical translation of gene therapy has been challenging in part because of the limitations of current delivery approaches. Herein, we report an efficient nonviral genome editor delivery approach using extracellular vesicles (EVs) carrying single-guide RNA (sgRNA): CRISPR-Cas9 ribonucleoprotein (RNP) complexes for in vivo gene therapy. By leveraging a high-throughput microfluidic droplet–based electroporation system (μDES), we achieved a 10-fold enhancement in loading efficiency and more than 1000-fold increase in processing throughput for loading RNP complexes into EVs compared with conventional high-voltage pulsed electroporation. μDES generated uniform microdroplets containing EVs and RNPs by applying direct current–controlled low voltage (up to 60 V) to transiently permeabilize membranes and enable efficient cargo encapsulation while maintaining EV integrity at both the protein and morphological levels. In the Myo7a WT/Sh1 mouse model of autosomal dominant progressive hearing loss, which may model MYO7A-associated DFNA11 hearing loss in humans, we demonstrated the effective delivery of RNPs by EVs into cochlear hair cells by cross-sectional and whole-mount confocal imaging. The injection of RNP-EVs via the posterior semicircular canal in 4-week-old Myo7a WT/Sh1 mice resulted in a reduction in Myo7a Sh1 messenger RNA expression and evidence of hearing preservation, as measured by auditory brainstem responses, compared with untreated ears and EV only–injected mice. This study highlights the potential of μDES-produced RNP-EVs for gene editing as a treatment for progressive nonsyndromic hearing loss in patients.
- 12 November 2025
- Published by American Association for the Advancement of Science (AAAS) in Science Translational Medicine
- Vol. 17 (824) , eadn3993
Extracellular vesicle–mediated gene editing for the treatment of nonsyndromic progressive hearing loss in adult mice
The clinical translation of gene therapy has been challenging in part because of the limitations of current delivery approaches. Herein, we report an efficient nonviral genome editor delivery approach using extracellular vesicles (EVs) carrying single-guide RNA (sgRNA): CRISPR-Cas9 ribonucleoprotein (RNP) complexes for in vivo gene therapy. By leveraging a high-throughput microfluidic droplet–based electroporation system (μDES), we achieved a 10-fold enhancement in loading efficiency and more than 1000-fold increase in processing throughput for loading RNP complexes into EVs compared with conventional high-voltage pulsed electroporation. μDES generated uniform microdroplets containing EVs and RNPs by applying direct current–controlled low voltage (up to 60 V) to transiently permeabilize membranes and enable efficient cargo encapsulation while maintaining EV integrity at both the protein and morphological levels. In the Myo7a WT/Sh1 mouse model of autosomal dominant progressive hearing loss, which may model MYO7A-associated DFNA11 hearing loss in humans, we demonstrated the effective delivery of RNPs by EVs into cochlear hair cells by cross-sectional and whole-mount confocal imaging. The injection of RNP-EVs via the posterior semicircular canal in 4-week-old Myo7a WT/Sh1 mice resulted in a reduction in Myo7a Sh1 messenger RNA expression and evidence of hearing preservation, as measured by auditory brainstem responses, compared with untreated ears and EV only–injected mice. This study highlights the potential of μDES-produced RNP-EVs for gene editing as a treatment for progressive nonsyndromic hearing loss in patients.
- Open Access
GatorAffinity: Boosting Protein-Ligand Binding Affinity Prediction with Large-Scale Synthetic Structural Data
Protein-ligand binding affinity prediction is a fundamental task in computational drug discovery. Although substantial efforts have been made to enhance prediction accuracy using data-driven approaches, progress remains limited by persistent data scarcity. The widely used PDBbind dataset, for example, contains fewer than 20,000 experimental structures with annotated binding affinities, while a vast number of affinity measurements remain underutilized due to missing structural data. Here, we investigate this untapped potential by curating more than 450,000 synthetic protein-ligand complexes annotated with Kd and Ki values using the Boltz-1 structure prediction model. Building on this unprecedented scale of synthetic data, further augmented with over 1 million synthetic complexes from the recently released SAIR database annotated with IC50 values, we develop GatorAffinity, a geometric deep learning-based scoring function pre-trained on large-scale synthetic data and fine-tuned using high-quality experimental structures from PDBbind. Extensive evaluation on a leak-proof benchmark demonstrates that GatorAffinity significantly outperforms state-of-the-art affinity prediction methods, offering superior accuracy and generalizability. Our findings show that augmenting available experimental data with synthetic complexes can effectively address the data scarcity challenge while maintaining strong predictive reliability. By releasing the pretrained GatorAffinity model and the large-scale synthetic dataset GatorAffinity-DB, we provide a scalable and reproducible foundation for affinity prediction, virtual screening, and broader structure-based drug design applications (https://github.com/AIDD-LiLab/GatorAffinity).
Open AccessGatorAffinity: Boosting Protein-Ligand Binding Affinity Prediction with Large-Scale Synthetic Structural Data
Protein-ligand binding affinity prediction is a fundamental task in computational drug discovery. Although substantial efforts have been made to enhance prediction accuracy using data-driven approaches, progress remains limited by persistent data scarcity. The widely used PDBbind dataset, for example, contains fewer than 20,000 experimental structures with annotated binding affinities, while a vast number of affinity measurements remain underutilized due to missing structural data. Here, we investigate this untapped potential by curating more than 450,000 synthetic protein-ligand complexes annotated with Kd and Ki values using the Boltz-1 structure prediction model. Building on this unprecedented scale of synthetic data, further augmented with over 1 million synthetic complexes from the recently released SAIR database annotated with IC50 values, we develop GatorAffinity, a geometric deep learning-based scoring function pre-trained on large-scale synthetic data and fine-tuned using high-quality experimental structures from PDBbind. Extensive evaluation on a leak-proof benchmark demonstrates that GatorAffinity significantly outperforms state-of-the-art affinity prediction methods, offering superior accuracy and generalizability. Our findings show that augmenting available experimental data with synthetic complexes can effectively address the data scarcity challenge while maintaining strong predictive reliability. By releasing the pretrained GatorAffinity model and the large-scale synthetic dataset GatorAffinity-DB, we provide a scalable and reproducible foundation for affinity prediction, virtual screening, and broader structure-based drug design applications (https://github.com/AIDD-LiLab/GatorAffinity).
- Open Access
- 28 September 2025
- Published by Scilight Press Pty Ltd in Transactions on Artificial Intelligence
- Vol. 1 (1) , 246-247
Perspectives on Artificial Intelligence for Medicine
Scilight Press is an international academic publisher dedicated to open-access peer-reviewed journals and books. As a global multidisciplinary publisher, we provide a platform for researchers, scholars, and academics to disseminate their work across diverse fields of study. We are committed to advancing knowledge and fostering academic discourse.
- 28 September 2025
- Published by Scilight Press Pty Ltd in Transactions on Artificial Intelligence
- Vol. 1 (1) , 246-247
Open AccessPerspectives on Artificial Intelligence for Medicine
Scilight Press is an international academic publisher dedicated to open-access peer-reviewed journals and books. As a global multidisciplinary publisher, we provide a platform for researchers, scholars, and academics to disseminate their work across diverse fields of study. We are committed to advancing knowledge and fostering academic discourse.