Drug Repurposing / Target Interaction on Vecura: Models, Methods & Use Cases
Introduction
The fastest path to a new medicine is often through an old one. Identifying which approved or investigational molecules bind a new target — and how tightly — is the central computational challenge of drug repurposing. On Vecura, scientists get one-click access to a full, integrated stack of drug-target interaction and repurposing tools — from ultrafast genome-scale DTI screening to physics-grade binding free energy calculations — without leaving the browser.
Background
Why It Matters
Drug repurposing — finding new therapeutic indications for compounds whose safety and pharmacokinetic profiles are already characterised — dramatically compresses development timelines and reduces attrition risk. Artificial intelligence now spans the full arc of this process, from protein structure prediction to knowledge-graph reasoning, accelerating candidate identification at every step [1]. Repurposing approved drugs reduces development cost and risk, and is an increasingly critical strategy for filling pipeline gaps — particularly against antimicrobial resistance — where conventional de novo discovery timelines are untenable [2]. Across the broader drug-discovery landscape, AI/ML approaches have curtailed candidate attrition by up to 30% and compressed timelines by 20–40%, with modern ML models predicting drug-target interactions at accuracies approaching ~85% [3]. Integration of multi-omics data with AI further accelerates target identification, drug repositioning, and de novo molecular design [4].
What It Is
Drug-target interaction (DTI) prediction is the computational estimation of whether — and how strongly — a small molecule or biologic binds a protein (or nucleic acid) target. DTI prediction underpins repurposing directly: running the same affinity machinery in reverse (one known drug screened against many targets, or many drugs screened against one target) surfaces new indications and flags off-target liabilities simultaneously. The methodological toolbox spans three complementary paradigms: ligand-based methods (chemical similarity, fingerprints), sequence-based methods (protein and molecular language-model embeddings that require no 3D structure), and structure-based methods (molecular docking and end-point or alchemical free-energy calculations). Natural-product target discovery is a canonical motivating example where the target is entirely unknown and reverse screening is the only entry point [5].
Drug Repurposing / Target Interaction Models on Vecura
Vecura's catalog spans every stage of the repurposing and target-interaction pipeline; the table below maps scientific use cases to the specific models available in-app and the rationale for each choice.
| Use Case | Model(s) | Why |
| Binding Affinity Prediction & Rescoring | ||
| Sequence-based affinity (pKd) from protein + ligand | BALM, DTIGN | Language-model / GNN affinity from sequence + SMILES — no 3D pose required; ideal for fast triage across large drug–target pair matrices |
| Structure-based affinity / pose rescoring | GEMS, PandaDock (score_protein_ligand_interaction) | GNN scoring on the 3D complex to re-rank docked poses; PandaDock achieves Pearson R ≈ 0.88 on PDBbind |
| Physics-based binding free energy | Uni-GBSA | MM/GB(PB)SA endpoint free energies for tight, physically grounded ranking of a short-listed compound set |
| Unsupervised binding energy | DSMBind | SE(3) denoising score-matching energy without labelled affinity data; covers protein–ligand and protein–protein interfaces |
| Drug-Target Interaction & Virtual Screening | ||
| Ultrafast DTI / large-library screening | SPRINT (panspecies-dti) | Co-embeds structure-aware protein representations and ligand fingerprints for genome- or library-scale DTI in a single pass |
| General DTI toolkit | DeepPurpose | Unified encoder-decoder framework supporting DTI, DDI, molecular property, and PPI prediction — a flexible, well-validated baseline |
| Ligand shape / fingerprint similarity screening | ROSHAMBO, FPSim2 | GPU Gaussian shape overlap (ROSHAMBO) and fast bit-vector fingerprint similarity (FPSim2) to identify repurposing analogs of a confirmed binder |
| Ultra-large combinatorial library search | Thompson Sampling for Virtual Screening | Active-learning bandit search over billions of enumerable products without exhaustive enumeration |
| Target Identification & Off-Target / Repurposing | ||
| Predict likely targets of a compound | LigTMap, ReverseLigQ | Reverse screening: LigTMap combines similarity + PSOVina docking + IF-ML scoring across 17 target classes; ReverseLigQ maps ligand → Pfam → protein to surface novel/off-targets |
| Binding-site / druggable-residue detection | AF2BIND | Pinpoints small-molecule-binding residues on a target structure without a co-crystallised ligand, enabling druggability assessment |
| Small-molecule–RNA interactions | SMRTnet | Extends target space to RNA, predicting binding from RNA secondary structure — important for emerging RNA-targeted repurposing campaigns |
| Docking (Pose Generation for Interaction Analysis) | ||
| Classical / CNN docking | AutoDock-Vina, gnina | Gold-standard search algorithm with optional CNN rescoring (gnina); produces poses suitable as input for any of the rescoring models above |
| Deep-learning / flexible docking | DiffDock, DynamicBind, SurfDock, FlowDock | Diffusion- and flow-based blind and flexible docking; DynamicBind explicitly models apo→holo protein conformational change; FlowDock additionally outputs a predicted affinity |
| Ligand / receptor preparation | Meeko, MolScrub | PDBQT conversion, 3D conformer generation, tautomer/protonation-state enumeration — essential for robust, reproducible docking inputs |
| Complex Structure + Affinity (Co-Folding) | ||
| Co-fold complex and predict affinity | Boltz-2, Boltz-2.1, AlphaFold3 | Jointly predict the 3D protein–ligand (or protein/nucleic acid) complex structure; Boltz-2/2.1 additionally output a predicted binding affinity — structure and score in one run |
| Drug-Drug Interaction & Combinations (Repurposing Safety) | ||
| Typed DDI prediction | DeepDDI2, DeepDDI, DeepPurpose (DDI) | Predict dozens to hundreds of human-readable DDI types to flag pharmacological combination risks when repositioning a compound into a new therapeutic context |
| Polypharmacy side effects | Decagon | Multi-relational GCN predicts side-effect types arising from specific drug pairs — critical for polypharmacy-heavy repurposing scenarios |
| Drug-combination synergy | SynergyFinder | HSA/Loewe/Bliss/ZIP synergy scoring for evaluating whether repurposed drug combinations are synergistic, additive, or antagonistic |
| Protein-Protein Affinity (Biologics Repurposing) | ||
| PPI binding affinity / ΔΔG | PRODIGY, PPAP, StaB-ddG | Contact-based (PRODIGY) and ESM2-based (PPAP, StaB-ddG) affinity and mutation ΔΔG prediction for protein–protein interfaces; supports biologics repurposing and engineering |
Notes
• Start with what you have. If you have only sequence and SMILES, begin with BALM or SPRINT. If you have a 3D complex (experimental or co-folded), proceed directly to GEMS, Uni-GBSA, or DSMBind for higher-fidelity scoring.
• Run a computational funnel. The recommended workflow is: cheap DTI/similarity screening (SPRINT, FPSim2) → docking for pose generation (Vina, DiffDock) → GNN/physics rescoring on a short-list (GEMS, Uni-GBSA). Each stage eliminates non-starters before committing compute.
• Predicted affinities are relative, not absolute. Rankings from any in silico model should be treated as prioritisation scores; experimental validation (binding assay, SPR, ITC) of top hits remains essential before advancing candidates.
• Chain models into Vecura Workflows. These modules compose natively — for example: AF2BIND (druggability) → SPRINT (library triage) → DiffDock (pose) → GEMS (rescore) → DeepDDI2 (safety check) — running end-to-end without manual file transfers.
• The catalog evolves. New models are added regularly; always consult the live model catalog in-app for the most current list, versioning, and parameter options.
Conclusion
Vecura transforms drug repurposing from a fragmented, multi-tool effort into a single connected environment where you can reason about target druggability, screen compound libraries, generate and score binding poses, and assess combination safety — all within the browser and all on the same data. Use the table above as your entry point: identify the use case that matches your current data and scientific question, select the appropriate model tier, and chain the outputs into a Workflow to move from hypothesis to ranked, safety-checked repurposing candidates in hours rather than weeks.
References
-
Artificial intelligence in drug research and development: a review of methods and applications in drug repurposing — https://doi.org/10.1093/bib/bbag203
-
Old drugs, new weapons: current trends in repurposing therapies against antimicrobial resistance — https://doi.org/10.3389/jpps.2026.16158
-
Advancing Drug Discovery with AI: Machine and Deep Learning Strategies for Target Identification and Precision Nanomedicine — https://doi.org/10.2147/ijn.s600651
-
Multi-omics and artificial intelligence for precision drug discovery and potential clinical applications — https://doi.org/10.1038/s41392-026-02631-6
-
Progress in Natural Products Target Discovery Technology — https://doi.org/10.1002/mco2.70777
Try Vecura now.
Bring your own inputs and start exploring what Vecura can do.
