The study, published in the journal Molecular Diversity, was led by Binglin Huang and Bijuan Lin of Fujian Medical University Union Hospital, together with colleagues at Fujian Medical University and Nanjing University of Chinese Medicine. Rather than relying on the well-characterized binding groove that VHL uses to recognize hydroxylated HIF-1α, the team took a different route. They used DDMut-PPI, a graph-based deep-learning tool that predicts how mutations at a protein-protein interface affect the interaction, to nominate residues that might serve as auxiliary contact points beyond the canonical binding site. Those predicted residues were then used to build an alternative pharmacophore model, a simplified three-dimensional map of the chemical features a molecule would need to wedge itself into the interface and interfere with the HIF-1α/VHL handshake.

With that model in hand, the researchers screened a library of 24,893 molecules computationally, using docking tools including AutoDock4, which allows selective receptor flexibility during the docking calculation. Candidates that docked favorably against the alternative pharmacophore were then carried into the laboratory, where a fluorescence-polarization assay measured how well each compound competed with a labeled HIF-1α peptide for binding to VHL. Four molecules emerged with half-maximal inhibitory concentrations, or IC50 values, below 10 micromolar, a respectable starting point for compounds discovered without any structure-guided medicinal chemistry.

The standout was a compound the authors labeled Cmpd16, which turned out to be ixazomib, a proteasome inhibitor already approved for use in multiple myeloma under the trade name Ninlaro. In the fluorescence-polarization assay, Cmpd16 showed the highest affinity of the four hits, with an IC50 of 0.41 micromolar. The finding that an existing drug can bind VHL and interfere with HIF-1α recognition is notable because it opens the possibility of repurposing: a molecule with known pharmacokinetics, safety data, and manufacturing routes could, in principle, be repositioned for a completely different indication, provided the new activity holds up in more demanding biological systems.

The team then asked whether the compound actually did what a VHL inhibitor should do inside cells. In their reported assays, Cmpd16 showed no detectable loss of cell viability, an important early safety signal, and it produced a VHL-dependent pattern of stabilization affecting both HIF-1α and the hydroxylated form of the protein. In other words, when VHL’s grip is loosened, even hydroxylated HIF-1α, which would normally be destroyed regardless of oxygen levels, lingers in the cell. That is precisely the biochemical signature expected from a molecule that blocks the VHL recognition step downstream of the oxygen-sensing hydroxylases, and it mirrors the mechanism of dedicated VHL inhibitors developed over the past decade by academic and industrial groups.

To test whether this molecular effect translated into protection under stress, the researchers used an oxygen-glucose deprivation and reoxygenation model, a standard laboratory mimic of ischemia and reperfusion injury in which cells are deprived of both oxygen and nutrients and then resupplied. At a concentration of 10 micromolar, Cmpd16 improved the migration of endothelial cells and their ability to form tube-like structures, processes central to the formation of new blood vessels that resupply damaged tissue. The treated cells also showed increased levels of vascular endothelial growth factor, VEGF, and the glucose transporter GLUT1, both of which are canonical HIF-1α target genes that support angiogenesis and metabolic adaptation. At the same time, the compound was associated with reduced accumulation of reactive oxygen species and reduced levels of cleaved caspase-3, a hallmark of apoptosis, suggesting that the cells were not only adapting but genuinely surviving the insult better.

Because the initial hit came from a computational model rather than from a crystal structure of a ligand bound in the interface, the authors took care to interrogate how plausible their predicted binding mode really was. They ran three independently initialized molecular dynamics simulations of 200 nanoseconds each using the Desmond engine. Across the replicas, the simulations showed recurring contacts between the ligand and two residues, Pro99 and His110, in VHL, but the overall dynamics of both the protein and the ligand varied from replica to replica. The authors are explicit about what this means: the simulations support a computationally plausible orientation for Cmpd16 rather than a unique, well-defined one. That honesty matters, because molecular dynamics can be seductive, and overstating a single binding pose from a handful of trajectories is a well-known pitfall in computational drug discovery.

The team also compared two selected ten-compound panels to probe whether their deep-learning-informed pharmacophore enriched for active molecules more effectively than a conventional approach. The comparison, analyzed with a two-sided Fisher’s exact test, yielded a p-value of 0.0867, which falls short of conventional statistical significance. The authors describe this analysis as exploratory, and they further caution that the causal role of the predicted interface residues remains unproven until experimental mutagenesis is performed. These caveats are worth emphasizing: the study demonstrates a viable discovery pipeline and a promising hit, but the mechanistic details of where and how the compound binds VHL are still hypotheses awaiting direct experimental confirmation.

The broader context makes the work interesting on several fronts. The HIF pathway has long been a drug-discovery target, though most clinical success has come from a different angle: inhibitors of the prolyl hydroxylase enzymes, such as daprodustat and enarodustat, which are approved or in late-stage development for anemia in chronic kidney disease. Those drugs prevent HIF-1α from being tagged in the first place. Directly blocking the VHL-HIF interaction is a more recent strategy, and potent chemical probes with nanomolar affinities have been reported, largely built around a conserved hydroxyproline mimetic scaffold. The structural diversity of known VHL ligands has remained narrow, which is exactly the gap the new study set out to address by looking for chemically distinct molecules through an alternative interface model. Finding an approved proteasome inhibitor within that chemically distinct set is an unexpected twist, since ixazomib was designed to inhibit the proteasome’s catalytic activity, not to bind an E3 ubiquitin ligase.

There is also a pleasing irony in the biology. VHL ligands were originally developed largely as tools for targeted protein degradation, forming the binding arm of PROTAC molecules that recruit VHL to destroy disease-causing proteins. Here, the same interaction is being blocked for the opposite purpose: to prevent VHL from destroying a protein the body needs during oxygen crisis. Whether Cmpd16, or molecules inspired by it, can eventually protect tissues in living models of ischemia remains to be seen, and the distance from an endothelial-cell dish to a patient’s heart or kidney is considerable. But the study offers a concrete example of how deep-learning predictions of interface residues can reshape the search space in virtual screening, and it adds a repurposing candidate, and a set of testable residue hypotheses, to a pathway whose pharmacological importance continues to grow. The next steps, mutagenesis of the predicted contact residues, co-crystallography or cryo-EM of the ligand bound to VHL, and validation in animal models of ischemic injury, will determine whether this computational detour leads to a genuinely new way of protecting cells when oxygen runs out.

Subject of Research: Computational discovery of small-molecule inhibitors of the HIF-1α/VHL protein-protein interaction for protection of cells under hypoxic stress

Article Title: Computational discovery of HIF-1α/VHL protein–protein interaction inhibitors for hypoxic cell protection

Article References: Computational discovery of HIF-1α/VHL protein–protein interaction inhibitors for hypoxic cell protection. (n.d.). https://doi.org/10.1007/s11030-026-11706-z

Image Credits: AI Generated

DOI: 10.1007/s11030-026-11706-z

Keywords: HIF-1α, VHL, protein-protein interaction, virtual screening, deep learning, ixazomib, hypoxia, ischemia, molecular dynamics, drug repurposing, endothelial cells, pharmacophore

The new work builds on a wave of geometric learning methods that have transformed computational docking in recent years. Classical tools such as AutoDock, AutoDock Vina, GOLD, and MedusaDock combine a conformational search over the ligand’s translational, rotational, and torsional degrees of freedom with scoring functions that approximate the energetics of binding. Machine learning scoring functions like NNScore and RF-Score later learned nonlinear relationships between structural features and binding properties, and deep models such as AtomNet and GNINA learned interaction representations directly from three-dimensional structures. More recently, methods like EquiBind, TANKBind, E3Bind, FABind, and DiffDock have moved beyond scoring toward direct prediction or refinement of ligand geometry, with DiffDock famously reformulating pose prediction as a diffusion process over translational, rotational, and torsional variables. Systems such as DynamicBind, NeuralPLexer, AlphaFold 3, FABFlex, and FlowDock have extended these ideas to settings involving protein flexibility and joint modeling of entire protein-ligand complexes.

Yet the authors of the new study identified a persistent gap in how refinement methods handle coordinate errors. A docked ligand may need a common shift in position and orientation, which is a rigid transformation applied to the whole molecule, while at the same time individual atoms may require their own local adjustments to match the true bound conformation. A global transformation alone cannot represent atom-specific corrections, and an atom-wise displacement parameterization, the approach used by the earlier graph-based method MedusaGraph, never explicitly parameterizes a shared rotation and translation for the complete ligand. Pose accuracy must also be considered alongside molecular geometry: a low coordinate error does not by itself guarantee that the ligand’s internal structure is preserved or that it avoids physically implausible overlap with the protein. These observations motivated a design that combines complementary coordinate predictions with pose supervision and additional geometric regularization.

KTransPose answers that challenge with a dual-branch architecture built on a graph neural network encoder. The protein-ligand complex is represented as a graph whose nodes are the ligand atoms and the selected protein-pocket atoms, with edges encoding molecular connectivity and interatomic distances. The encoder, which uses a hidden dimension of 256 and residual connections with GELU activations and dropout, produces a shared representation of the graph. From this representation, a global pose transformation branch pools the ligand and protein node features separately, concatenates them, and passes them through a small multilayer perceptron that outputs an axis-angle rotation vector and a translation vector. The rotation is converted to a valid rotation matrix using Rodrigues’ formula, and the transformation is applied around the ligand centroid. Meanwhile, a local coordinate correction branch applies an additional graph layer with three output channels, producing an independent three-dimensional adjustment for every ligand atom. The final displacement for each atom is simply the sum of the global and local contributions.

The training objective is where the framework earns its docking-oriented character. Rather than supervising displacements directly, the loss operates on the predicted absolute coordinates and combines four components. The first is the root mean squared distance computed in the original receptor coordinate frame, without any alignment, which directly supervises where the ligand sits in the binding site. The second is a mean squared error computed after optimally aligning the prediction to the reference using the classical Kabsch algorithm, implemented differentiably so that gradients flow through the singular value decomposition; this auxiliary term evaluates structural agreement after removing the best rigid transformation. The third term regularizes the complete pairwise distance matrices of the predicted and reference ligand structures, preserving internal ligand geometry independently of absolute position. The fourth is a clash penalty that quadratically punishes protein-ligand atom pairs closer than a cutoff of 2.0 in the normalized coordinate representation. A sensitivity analysis settled on weights of 1.0, 0.10, 0.05, and 0.02 for these four terms, respectively.

That sensitivity analysis produced one of the study’s most instructive findings. With only the receptor-frame RMSD term active, the dual-branch model achieved a mean RMSD of 5.08 angstroms on the PDBbind-2020 benchmark. Adding the auxiliary terms lowered this to 4.67 angstroms, a clear demonstration that alignment-aware, intraligand, and clash supervision genuinely help. But the reverse experiment was even more striking: when the weight of the raw RMSD term was reduced while the auxiliary terms stayed fixed, performance collapsed catastrophically, from 4.67 angstroms at full weight to 27.96 angstroms when the term was removed entirely. The auxiliary geometric terms complement rather than replace the supervision needed for correct absolute placement, a nuance the authors argue is essential for anyone designing docking losses.

The team evaluated seven graph neural network configurations within the framework, including TransformerConv, GraphSAGE, Graph Attention Networks, ARMAConv, and three hybrid architectures combining GIN, GAT, and TransformerConv layers. TransformerConv, an attention-based message-passing operator that can incorporate edge attributes, delivered the strongest overall performance, followed by GraphSAGE and the G3 hybrid, though it also required the most training time per epoch. All experiments were run on modest hardware, an Intel Core i7-7700 desktop with a GeForce GTX 1080 GPU, and five training runs with different random seeds showed a standard deviation of only about 0.01 angstroms, indicating that the results are highly reproducible.

The headline comparison, against MedusaGraph under strictly identical conditions, same initial MedusaDock poses, same preprocessing, same test complexes, atom ordering, graph construction, and RMSD calculation, showed consistent gains. On PDBbind-2020, a dataset of 3,956 preprocessed protein-ligand complexes split by CD-HIT sequence clustering at 90 percent identity, MedusaGraph achieved a mean RMSD of 5.08 angstroms, while KTransPose reached 4.67, 4.51, and 4.47 angstroms across three successive refinement iterations. On the independent CASF-2016 benchmark of 278 evaluated complexes, MedusaGraph’s 4.81 angstroms fell to 4.41, 4.27, and 4.24 angstroms for KTransPose, an improvement of roughly 12 percent on both datasets. Paired two-sided Wilcoxon signed-rank tests confirmed the differences were statistically significant at the matched iterations. Notably, iterative refinement for KTransPose reduced RMSD monotonically, whereas MedusaGraph actually degraded at its third iteration on both datasets, rising from 4.68 to 4.90 angstroms on PDBbind-2020.

The refinement dynamics themselves carry a practical lesson. Most of the total improvement over the initial MedusaDock pose is achieved in the very first iteration, with the second and third iterations contributing progressively smaller gains, just 0.16 and then 0.04 angstroms for the TransformerConv model on PDBbind-2020. Because each additional iteration requires applying the preceding frozen models, updating coordinates, and reconstructing the ligand-dependent graph edges within a 6 angstrom distance threshold, the authors stopped at three iterations, judging the diminishing returns not worth the extra computational cost. Success-rate analyses at fixed RMSD thresholds added further texture: KTransPose showed its clearest advantages at the 3 and 5 angstrom cutoffs, while the strict 2 angstrom success rate remained low for both methods and was not uniformly improved. A supplementary analysis using the DAAP affinity predictor also found that KTransPose-refined complexes yielded better affinity-prediction metrics across Pearson correlation, error, and concordance measures than MedusaGraph outputs.

The authors are candid about the boundaries of their results. KTransPose refines an existing docked pose rather than performing blind or de novo docking, so its performance depends on the quality of the supplied MedusaDock starting point, and protein coordinates are held fixed throughout, meaning receptor flexibility and conformational change are not modeled. The comparison with MedusaGraph is a method-level comparison, so the gains cannot be attributed to the global branch or the loss function alone, and the evaluation rests primarily on PDBbind-2020 with CASF-2016 as an additional test. Even so, the framework’s consistent improvements, its reproducibility, and its publicly released codebase mark it as a meaningful step for computational drug discovery, and the authors point toward equivariant architectures, flexible-receptor settings, and more diverse docking initializations as the next frontiers. For a field where fractions of an angstrom can separate a plausible drug candidate from a dead end, a reliable 12 percent reduction in pose error is no small matter.

Subject of Research: Deep learning-based refinement of protein-ligand binding pose prediction in structure-based drug discovery

Article Title: Enhancing protein–ligand pose prediction via dual-branch deep learning with docking-oriented multi-component loss and iterative refinement

Article References: Alam, M. K., Rahman, J., Newton, M. A. H., & Ali, M. E. (2026). Enhancing protein–ligand pose prediction via dual-branch deep learning with docking-oriented multi-component loss and iterative refinement. Applied Intelligence, 56(15), Article 458. https://doi.org/10.1007/s10489-026-07475-9

Image Credits: AI Generated

DOI: 10.1007/s10489-026-07475-9

Keywords: protein-ligand pose prediction, molecular docking, graph neural networks, drug discovery, KTransPose, MedusaGraph, PDBbind-2020, CASF-2016, Kabsch alignment, TransformerConv, iterative refinement, virtual screening