Deep learning for protein structure prediction and design—progress and applications

Molecular Systems Biology, Journal Year: 2024, Volume and Issue: 20(3), P. 162 - 169

Published: Jan. 30, 2024

Abstract Proteins are the key molecular machines that orchestrate all biological processes of cell. Most proteins fold into three-dimensional shapes critical for their function. Studying 3D shape can inform us mechanisms underlie in living cells and have practical applications study disease mutations or discovery novel drug treatments. Here, we review progress made sequence-based prediction protein structures with a focus on go beyond single monomer structures. This includes application deep learning methods complexes, different conformations, evolution these to design. These developments create new opportunities research will impact across many areas biomedical research.

Language: Английский

---

Advances, opportunities, and challenges in methods for interrogating the structure activity relationships of natural products

Natural Product Reports, Journal Year: 2024, Volume and Issue: 41(10), P. 1543 - 1578

Published: Jan. 1, 2024

This review highlights methods for studying structure activity relationships of natural products and proposes that these are complementary could be used to build an iterative computational-experimental workflow.

Language: Английский

---

Comparative Structure-Based Virtual Screening Utilizing Optimized AlphaFold Model Identifies Selective HDAC11 Inhibitor

International Journal of Molecular Sciences, Journal Year: 2024, Volume and Issue: 25(2), P. 1358 - 1358

Published: Jan. 22, 2024

HDAC11 is a class IV histone deacylase with no crystal structure reported so far. The catalytic domain of shares low sequence identity other HDAC isoforms, which makes conventional homology modeling less reliable. AlphaFold machine learning approach that can predict the 3D proteins high accuracy even in absence similar structures. However, fact models are predicted small molecules and ions/cofactors complicates their utilization for drug design. Previously, we optimized an model by adding zinc ion minimization presence inhibitors. In current study, implement comparative structure-based virtual screening utilizing previously to identify novel selective stepwise was successful identifying hit subsequently tested using vitro enzymatic assay. compound showed IC50 value 3.5 µM could selectively inhibit over subtypes at 10 concentration. addition, carried out molecular dynamics simulations further confirm binding hypothesis obtained docking study. These results reinforce presented optimization applicability search inhibitors discovery.

Language: Английский

---

Learnt representations of proteins can be used for accurate prediction of small molecule binding sites on experimentally determined and predicted protein structures

Journal of Cheminformatics, Journal Year: 2024, Volume and Issue: 16(1)

Published: March 14, 2024

Protein-ligand binding site prediction is a useful tool for understanding the functional behaviour and potential drug-target interactions of novel protein interest. However, most methods are tested by providing crystallised ligand-bound (holo) structures as input. This testing regime insufficient to understand performance on targets where experimental not available. An alternative option provide computationally predicted structures, but this commonly tested. due training data used, computationally-predicted tend be extremely accurate, often biased toward holo conformation. In study we describe benchmark IF-SitePred, protein-ligand method which based labelling ESM-IF1 language model embeddings combined with point cloud annotation clustering. We show that only IF-SitePred competitive state-of-the-art when predicting sites it performs better proxies proteins low accuracy has been simulated molecular dynamics. Finally, outperforms other if ensembles generated.

Language: Английский

---

AlphaFold2 Predicts Whether Proteins Interact Amidst Confounding Structural Compatibility

Journal of Chemical Information and Modeling, Journal Year: 2024, Volume and Issue: 64(5), P. 1473 - 1480

Published: Feb. 19, 2024

Predicting whether two proteins physically interact is one of the holy grails computational biology, galvanized by rapid advancements in deep learning. AlphaFold2, although not developed with this goal, promising respect. Here, I test prediction capability AlphaFold2 on a very challenging data set, where are structurally compatible, even when they do interact. achieves high discrimination between interacting and non-interacting proteins, cases misclassifications can either be rescued revisiting input sequences or suggest false positives negatives set. thus impaired compatibility protein structures has potential to applied large scale.

Language: Английский

---

DiffBindFR: an SE(3) equivariant network for flexible protein–ligand docking

Chemical Science, Journal Year: 2024, Volume and Issue: 15(21), P. 7926 - 7942

Published: Jan. 1, 2024

DiffBindFR, a diffusion model based flexible full-atom protein–ligand docking tool, demonstrates its superior and side-chain refinement accuracy with reliable physical plausibility.

Language: Английский

---

Reliability of AlphaFold2 Models in Virtual Drug Screening: A Focus on Selected Class A GPCRs

International Journal of Molecular Sciences, Journal Year: 2024, Volume and Issue: 25(18), P. 10139 - 10139

Published: Sept. 21, 2024

Protein three-dimensional (3D) structure prediction is one of the most challenging issues in field computational biochemistry, which has overwhelmed scientists for almost half a century. A significant breakthrough structural biology been established by developing artificial intelligence (AI) system AlphaFold2 (AF2). The AF2 provides state-of-the-art protein structures from nearly all known sequences with high accuracy. This study examined reliability models compared to experimental drug discovery, focusing on common drug-targeted classes as G protein-coupled receptors (GPCRs) class A. total 32 representative targets were selected, including X-ray crystallographic and Cryo-EM their corresponding models. quality was assessed using different validation tools, pLDDT score, RMSD value, MolProbity percentage Ramachandran favored, QMEAN Z-score, QMEANDisCo Global. molecular docking performed Genetic Optimization Ligand Docking (GOLD) software. models’ virtual screening determined ability predict ligand binding poses closest native pose assessing Root Mean Square Deviation (RMSD) metric scoring function. function evaluated enrichment factor (EF). Furthermore, capability identify hits key protein–ligand interactions analyzed. posing power results showed that successfully predicted (RMSD < 2 Å). However, they exhibited lower power, average EF values 2.24, 2.42, 1.82 X-ray, Cryo-EM, structures, respectively. Moreover, our revealed can competitive inhibitors. In conclusion, this found provided comparable particularly certain GPCR targets, could potentially significantly impact discovery.

Language: Английский

---

Proteome-Wide Identification and Comparison of Drug Pockets for Discovering New Drug Indications and Side Effects

Molecules, Journal Year: 2025, Volume and Issue: 30(2), P. 260 - 260

Published: Jan. 10, 2025

Drug development faces significant financial and time challenges, highlighting the need for more efficient strategies. This study evaluated druggability of entire human proteome using Fpocket. We identified 15,043 druggable pockets in 20,255 predicted protein structures, significantly expanding estimated from 3000 to over 11,000 proteins. Notably, many were found less studied proteins, suggesting untapped therapeutic opportunities. The results a pairwise pocket similarity analysis 220,312 similar pairs, with 3241 pairs across different families, indicating shared drug-binding potential. In addition, 62,077 matches between 1872 known drug pockets, candidates repositioning. repositioned progesterone ADGRD1 pemphigus breast cancer, as well estradiol ANO2 shingles medulloblastoma, which validated via molecular docking. Off-target effects analyzed assess safety drugs such axitinib, linking newly targets side effects. For 127 new identified, 46 out 48 documented linked these targets. These findings demonstrate utility repositioning, target expansion, improved evaluation, offering avenues discovery indications existing drugs.

Language: Английский

---

Advancements in protein structure prediction: A comparative overview of AlphaFold and its derivatives

Computers in Biology and Medicine, Journal Year: 2025, Volume and Issue: 188, P. 109842 - 109842

Published: Feb. 18, 2025

Language: Английский

---

AlphaFold2 structures template ligand discovery

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2023, Volume and Issue: unknown

Published: Dec. 21, 2023

AlphaFold2 (AF2) and RosettaFold have greatly expanded the number of structures available for structure-based ligand discovery, even though retrospective studies cast doubt on their direct usefulness that goal. Here, we tested unrefined AF2 models prospectively, comparing experimental hit-rates affinities from large library docking against vs same screens targeting receptors. In σ2 5-HT2A receptors, struggled to recapitulate ligands had previously found receptors' structures, consistent with published results. Prospective models, however, yielded similar hit rates both receptors versus experimentally-derived structures; hundreds molecules were prioritized each model structure receptor. The success was achieved despite differences in orthosteric pocket residue conformations targets structures. Intriguingly, receptor most potent, subtype-selective agonists discovered via model, not structure. To understand this a molecular perspective, cryoEM determined one more potent selective emerge Our findings suggest may sample are relevant much extending domain applicability discovery.

Language: Английский

---