The challenge of achieving accurate binding affinity predictions due to the dependence on various parameters

The gap

Evidence profile

Sourced from the future work and stated challenges of the source papers, classified as general, drawn from work published between 2024 and 2026, spanning 4 journals. Those papers have been cited 289 times in total.

Supporting evidence — 6 representative gaps

• Structure and dynamics in drug discovery (2024) · npj Drug Discovery. · cited 65× · doi

  In the last decades, with all the successes and challenges, the transformation from computer-aided drug discovery to computer-driven drug discovery is emerging. The rapid advancements in structural biology along with the bloom of computational protein-structure prediction, are allowing for access to many more high-resolution 3D structures of novel drug-receptor complexes. Dynamic-based drug discovery strategies, coupled with advanced sampling methods, are offering new possibilities for drug design. During the early hit identification stage, the ultra-scale virtual screening approaches, both structure-based and AI-based, are becoming ready in providing fast and cost-effective start points into drug discovery campaigns. The rapid expansion of accessible chemical space, together with the dra- matic increases in computational power such as GPUs and cloud com- puting, are resulting in the ability to virtually screen multi-billions of drug- like chemical space. At the hit-to-lead stage, the more elaborate potency prediction tools such as free energy perturbation and AI-based QSAR often guide rational optimization of ligand potency. In the end, data-driven computational tools are used for multi-parameter optimization, including solubility, permeability, and pharmacokinetic properties, to identify the final drug candidate. It is evident that these breakthroughs are converging towards a new era of computer-driven drug discovery (Fig. 4).

  generalfuture work
  Keywords: drug discovery based computer driven computational rapid structure prediction stage chemical space multi potency tools
• Structure and dynamics in drug discovery (2024) · npj Drug Discovery. · cited 65× · doi

  AlphaFold 3 claim in their publication that the performance of their models on protein-ligand systems is better than classical docking tools, such as Vina97,98, and greatly outperforms all other blind dockings like RoseTTA- Fold All-Atom. Their evaluation was done on their PoseBusters benchmark set, which is composed of 428 protein-ligand structures that were not included in their training. The reported accuracy, as the percentage of protein-ligand pairs with pocket-aligned ligand root mean squared devia- tion of less than 2 Å, is over 90% for their high-confidence group. AlphaFold and other Machine learning models may be limited in the near future in their ability to predict cryptic pockets, given limitations in training data. It may be possible to create training data by enhanced sam- pling MD and use this data to train machine learning models to produce additional target structures. Related research is in Lyu, etc.99, where the authors followed up on several hundred computational hits and found that there was little to no overlap for the same receptor when starting with the AlphaFold 2 model versus the experimental structure. This indicated that AlphaFold models are already showing some potential on modeling dif- ferent conformations.

  generalfuture work
  Keywords: alphafold models ligand protein training structures machine learning claim publication performance systems better classical docking
• Structure and dynamics in drug discovery (2024) · npj Drug Discovery. · cited 65× · doi

  Future directions In the last decades, with all the successes and challenges, the transformation from computer-aided drug discovery to computer-driven drug discovery is emerging. The rapid advancements in structural biology along with the bloom of computational protein-structure prediction, are allowing for access to many more high-resolution 3D structures of novel drug-receptor complexes. Dynamic-based drug discovery strategies, coupled with advanced sampling methods, are offering new possibilities for drug design. During the early hit identification stage, the ultra-scale virtual screening approaches, both structure-based and AI-based, are becoming ready in providing fast and cost-effective start points into drug discovery campaigns. The rapid expansion of accessible chemical space, together with the dra- matic increases in computational power such as GPUs and cloud com- puting, are resulting in the ability to virtually screen multi-billions of drug- like chemical space. At the hit-to-lead stage, the more elaborate potency prediction tools such as free energy perturbation and AI-based QSAR often guide rational optimization of ligand potency. In the end, data-driven computational tools are used for multi-parameter optimization, including solubility, permeability, and pharmacokinetic properties, to identify the final drug candidate. It is evident that these breakthroughs are converging towards a new era of computer-driven drug discovery (Fig. 4).

  generalfuture work
  Keywords: drug discovery based computer driven computational rapid structure prediction stage chemical space multi potency tools
• AlphaFold3 versus experimental structures: assessment of the accuracy in ligand-bound G protein-coupled receptors (2024) · Acta Pharmacologica Sinica · cited 81× · doi

  Understanding ligand-protein interactions is essential for advancing drug discovery, protein design, and structural biology. Therefore, modeling ligand-protein interactions is a cornerstone of modern drug discovery and molecular biology, enabling the rational design of therapeutics and a deeper under- standing of biological processes [59]. Current commercial tools, such as AlphaFold3, have established a high standard for predicting the 3D structures of biomolecular complexes [6, 60]. However, their ac- cessibility and high costs limit broader adoption. The recently introduced fully open-source model, such as Boltz-1, addresses this issue and competes with these state-of-the-art tools in both accuracy and usability. We found that Boltz-1 demonstrated strong performance in reproducing protein folding, deriving its capabilities from the parent source code of AlphaFold 3. Boltz-1 effectively reproduces the 3D structures of biomolecular complexes, demonstrating excellent performance, particularly in docking ligands with flexible macrocycles. It successfully re-docked a diverse set of ligands with varying com- plexities, achieving binding scores comparable to commercial tools such as Glide by Schrodinger. In terms of RMSD ligand-binding ranking, Botlz-1 outperformed the popular docking tool AutoDock Vina for all ligands studied. Finally, while capturing the binding modes of low-molecular-weight or - ganics, predicting complex peptidomimetics remains beyond the current capabilities of Boltz-1. To summarize, our benchmarking indicates that Botlz-1 presents opportunities to improve computational 66 M. V. Prud, A. Kyrychenko screening of small molecular libraries and may play a significant role in the future of AI-driven pre- dicting of ligand-protein interactions. Finally, when our manuscript was ready for submission, Boltz’s developing team announced the revolutionizing update for Boltz-2 (https://github.com/jwohlwend/boltz), introducing controllability features including experimental method conditioning, distance constraints, and multi-chain template integration for structure prediction, and the AI model to approach the performance of free-energy per- turbation (FEP) methods in estimating small molecule–protein binding affinity. A.V.K. acknowledges the grant № 87/0062 (2021.01/0062) “Molecular design, synthesis and screening of new potential antiviral pharmaceutical ingredients for the treatment of infec- tious diseases COVID-19” from the National Research Foundation of Ukraine.

  generalfuture work
  Keywords: boltz protein ligand molecular binding interactions design tools performance ligands drug discovery biology current commercial
• Docking in the Dark: Insights into Protein–Protein and Protein–Ligand Blind Docking (2025) · Pharmaceuticals · cited 13× · doi

  4.1. Advances Over the past two decades, blind docking has evolved from rigid-body, geometry- based algorithms to data-driven and hybrid AI-enhanced frameworks that combine phys- ical interpretability with predictive power. Early approaches such as Hex, ZDOCK, and PatchDock provided the computational foundation for large-scale docking by emphasiz- ing geometric complementarity and speed, though they lacked flexibility and struggled to model dynamic protein behavior. The subsequent integration of energy-based poten- tials and reduced-representation models—exemplified by ATTRACT and FRODOCK— marked a transition toward physically informed methods that improved accuracy at man- ageable computational costs. Later, hybrid tools such as HADDOCK, SwarmDock, and Pharmaceuticals 2025, 18, 1777 14 of 18 pepATTRACT further advanced predictive performance by combining empirical scoring with interface prediction and post-docking refinement, effectively bridging the gap be- tween blind and guided docking. In the realm of protein–peptide docking, the methodological focus shifted toward capturing conformational flexibility and biological realism. Frameworks such as DynaDock, FlexPepDock, and HADDOCK introduced ensemble and dynamic-based sim- ulations that more accurately represented peptide adaptability. Subsequent innovations, including pepATTRACT, MDockPeP, and PatchMAN, demonstrated how coarse-grained modeling, hierarchical refinement, and motif-based recognition could collectively en- hance both computational efficiency and structural fidelity. This progression underscores a clear methodological trajectory—from empirical modeling of flexibility to predictive frameworks capable of generalizing across diverse peptide–protein systems. For ligand–protein blind docking, the field’s development mirrors a steady conver- gence between physics-based models and ML (ML) architectures. The initial reliance on search algorithms and scoring functions, as seen in AutoDock and EADock, gradually gave way to hybrid frameworks that integrate cavity detection, ensemble sampling, and statistical learning, such as CB-Dock, DockTScore, and EDock. The recent advent of DL (DL) and generative models—including DeepDock, EQUIBIND, TANKBind, DiffDock, and E3Bind—has further accelerated docking by predicting poses and binding sites with reduced computational cost. However, recent benchmark analyses (e.g., PoseBusters [66]) reveal that DL methods, while fast and scalable, still fall short in chemical realism and stereochemical accuracy compared to traditional physics-based approaches. The emer- gence of consensus and hybrid systems such as CoBDock [8] demonstrates the growing recognition that the optimal path forward lies in integrating the interpretability of physi- cal modeling with the adaptability of AI-driven prediction. 4.2. Challenges Despite these remarkable advancements, several fundam

  generalfuture work
  Keywords: docking based hybrid frameworks computational protein blind predictive exibility models peptide modeling algorithms driven interpretability
• Evaluating molecular docking for binding affinity predictions: a systematic analysis of key parameters and the utility of AlphaFold2 structures for the Schrödinger dataset (2026) · Journal of Computer-Aided Molecular Design · doi

  The study identifies the challenge of achieving accurate binding affinity predictions due to the dependence on various parameters. The research highlights the need for efficient downstream calculations, which depends on the quality of the starting poses. The paper notes the challenge of evaluating a large number of docking calculations, resulting in a prohibitively large number of permutations.

  generalstated challengesevidence 5/5
  Keywords: study identifies challenge achieving accurate binding affinity predictions

Questions about this gap