Benchmarking Refined and Unrefined AlphaFold2 Structures for Hit Discovery

Zhang, Yuqi; Vass, Márton; Shi, Da; Abualrous, Esam T.; Chambers, Jennifer M.; Chopra, Nikita; Higgs, Christopher; Kasavajhala, Koushik; Li, Hubert; Nandekar, Prajwal P.; Sato, Hideyuki; Miller, Edward B.; Repasky, Matthew P.; Jerome, Steven V.

doi:10.1021/acs.jcim.2c01219

Cited by 39 publications

(48 citation statements)

References 54 publications

(86 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The reduced performance of AF2 apo models when compared to X-ray holo structures is consistent with the recent studies of Schrodinger 22 and AutoDock developers, 34 and leads us to conclude that performing a rigid receptor docking exercise using out-of-the-box AF2 models is not an ideal scenario. A postprocessing step, including some pocket optimization, seems necessary to reach practically useful EF values.…”

Section: ■ Results and Discussionsupporting

confidence: 87%

“…20,21 Thus, it is important to assess the performance of AF2 structures when combined with current molecular modeling methods, an aspect that has already gained significant attention elsewhere. 22 For example, in a recent publication, we showed how a stateof-the-art implementation of FEP (FEP+ from Schrodinger), when applied to AF2 structures, could produce analogous results to those when using crystal structures. 23 In that study, in order to impose more realistic prospective conditions on our benchmark experiment, we developed a custom AF2 version, named AF2 30 , where we eliminated all structural templates with >30% identity to the target protein to make the structure prediction.…”

Section: ■ Introductionmentioning

confidence: 94%

“…Large differences between models may indicate that AF2 is gathering divergent evolutionary or structural data, producing structures that might not be reliable enough to be applied in DD campaigns. For the next step, before a large-scale virtual screening campaign, it would be ideal to run some induced-fit applications (such as PELE 37 or IFD-MD 22 ) using known inhibitors, aiming for a partial opening/accommodation of the active site residues. If the BS still appears to be collapsed, we recommend running extensive MD simulations, clustering, and Markov-State modeling to potentially identify stable conformations of the protein with open pockets capable of accommodating known ligands.…”

Section: ■ Conclusionmentioning

confidence: 99%

“…Similar results, when looking at enrichment factors (EFs) in docking calculations, have recently been provided by Schrodinger when using their latest induced fit approach (IFD-MD). 22 Importantly, EFs are reduced considerably, compared to those obtained with IFD-MD, when applying simple rigid docking tools (e.g., Glide) to AF2 structures.…”

Section: ■ Introductionmentioning

confidence: 99%

“…In this work, we provide an analysis, similar to that provided by Schrodinger, 22 of the use of AF2 models for docking calculations and virtual screening purposes. However, we attempt to simulate a more real-world scenario around carrying out virtual screening for a novel protein (i.e., a protein for which no related protein structures are available) by removing highidentity templates (those with identity >30%) prior to model building.…”

Section: ■ Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Are Deep Learning Structural Models Sufficiently Accurate for Virtual Screening? Application of Docking Algorithms to AlphaFold2 Predicted Structures

Díaz-Rovira

Martín²,

Beuming³

et al. 2023

J. Chem. Inf. Model.

View full text Add to dashboard Cite

Machine learning-based protein structure prediction algorithms, such as RosettaFold and AlphaFold2, have greatly impacted the structural biology field, arousing a fair amount of discussion around their potential role in drug discovery. While there are few preliminary studies addressing the usage of these models in virtual screening, none of them focus on the prospect of hit-finding in a real-world virtual screen with a model based on low prior structural information. In order to address this, we have developed an AlphaFold2 version where we exclude all structural templates with more than 30% sequence identity from the modelbuilding process. In a previous study, we used those models in conjunction with state-of-the-art free energy perturbation methods and demonstrated that it is possible to obtain quantitatively accurate results. In this work, we focus on using these structures in rigid receptor-ligand docking studies. Our results indicate that using out-of-the-box Alphafold2 models is not an ideal scenario for virtual screening campaigns; in fact, we strongly recommend to include some post-processing modeling to drive the binding site into a more realistic holo model.

show abstract

Section: ■ Results and Discussionsupporting

confidence: 87%

Section: ■ Introductionmentioning

confidence: 94%

Section: ■ Conclusionmentioning

confidence: 99%

Section: ■ Introductionmentioning

confidence: 99%

Section: ■ Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Are Deep Learning Structural Models Sufficiently Accurate for Virtual Screening? Application of Docking Algorithms to AlphaFold2 Predicted Structures

Díaz-Rovira

Martín²,

Beuming³

et al. 2023

J. Chem. Inf. Model.

View full text Add to dashboard Cite

show abstract

zPoseScore model for accurate and robust protein–ligand docking pose scoring in CASP15

Shen,

Liu,

Wang

et al. 2023

Proteins

View full text Add to dashboard Cite

We introduce a deep learning‐based ligand pose scoring model called zPoseScore for predicting protein–ligand complexes in the 15th Critical Assessment of Protein Structure Prediction (CASP15). Our contributions are threefold: first, we generate six training and evaluation data sets by employing advanced data augmentation and sampling methods. Second, we redesign the “zFormer” module, inspired by AlphaFold2's Evoformer, to efficiently describe protein–ligand interactions. This module enables the extraction of protein–ligand paired features that lead to accurate predictions. Finally, we develop the zPoseScore framework with zFormer for scoring and ranking ligand poses, allowing for atomic‐level protein–ligand feature encoding and fusion to output refined ligand poses and ligand per‐atom deviations. Our results demonstrate excellent performance on various testing data sets, achieving Pearson's correlation = 0.783 and 0.659 for ranking docking decoys generated based on experimental and predicted protein structures of CASF‐2016 protein–ligand complexes. Additionally, we obtain an averaged local distance difference test (lDDT pli = 0.558) of AIchemy LIG2 in CASP15 for de novo protein–ligand complex structure predictions. Detailed analysis shows that accurate ligand binding site prediction and side‐chain orientation are crucial for achieving better prediction performance. Our proposed model is one of the most accurate protein–ligand pose prediction models and could serve as a valuable tool in small molecule drug discovery.

show abstract

Accurate ligand–protein docking in CASP15 using the ClusPro LigTBM server

Kotelnikov,

Ashizawa,

Popov

et al. 2023

Proteins

View full text Add to dashboard Cite

In the ligand prediction category of CASP15, the challenge was to predict the positions and conformations of small molecules binding to proteins that were provided as amino acid sequences or as models generated by the AlphaFold2 program. For most targets, we used our template‐based ligand docking program ClusPro ligTBM, also implemented as a public server available at https://ligtbm.cluspro.org/. Since many targets had multiple chains and a number of ligands, several templates, and some manual interventions were required. In a few cases, no templates were found, and we had to use direct docking using the Glide program. Nevertheless, ligTBM was shown to be a very useful tool, and by any ranking criteria, our group was ranked among the top five best‐performing teams. In fact, all the best groups used template‐based docking methods. Thus, it appears that the AlphaFold2‐generated models, despite the high accuracy of the predicted backbone, have local differences from the x‐ray structure that make the use of direct docking methods more challenging. The results of CASP15 confirm that this limitation can be frequently overcome by homology‐based docking.

show abstract

Benchmarking Refined and Unrefined AlphaFold2 Structures for Hit Discovery

Cited by 39 publications

References 54 publications

Are Deep Learning Structural Models Sufficiently Accurate for Virtual Screening? Application of Docking Algorithms to AlphaFold2 Predicted Structures

Are Deep Learning Structural Models Sufficiently Accurate for Virtual Screening? Application of Docking Algorithms to AlphaFold2 Predicted Structures

zPoseScore model for accurate and robust protein–ligand docking pose scoring in CASP15

Accurate ligand–protein docking in CASP15 using the ClusPro LigTBM server

Contact Info

Product

Resources

About