The power and pitfalls of AlphaFold2 for structure prediction beyond rigid globular proteins

“…with its deep learning algorithm that enables accurate prediction of three-dimensional (3D) protein structures from only the target amino acid sequence, potentially solving the half-a-century-old protein structure-prediction problem: how to predict 3D structures from only sequence information 1–4 .”

“For targets without known homologs, such as computationally designed proteins, increasing the Perspective https://doi.org/10.1038/s41589-024-01638-w number of recycling iterations can improve the quality of the prediction 8 .”

“Several efforts have alleviated these limitations, including the Alpha-Fold Protein Structure Database 6,7 , which houses over 200 million pre-run AF2 predictions, as well as ColabFold 8 and OpenFold 9 , which allow users to run a modified AF2 protocol on open-access servers in minutes.”

“When more than one input sequence is provided, AlphaFold-Multimer or AF2Complex is used 20,21 .”

“The authors of AF2 note that, because it is trained on both apo and holo structures from the PDB, models may still be consistent with the expected structure in the presence of ligands or cofactors despite their absence in the AF2 workflow 1,5 .”

“A recent adaption of AlphaFold, AlphaMissense, does not explicitly determine the structural effects of a mutation on a protein but provides the probability of a missense variant being pathogenic 51 .”

“Several efforts have alleviated these limitations, including the Alpha-Fold Protein Structure Database 6,7 , which houses over 200 million pre-run AF2 predictions, as well as ColabFold 8 and OpenFold 9 , which allow users to run a modified AF2 protocol on open-access servers in minutes.”

“A separate study by Bryant et al. 56 reported a 63% success rate for heterodimeric complexes.”

“Typically, χ 2 values of less than one are indicative of overfitting, and additional strategies such as the combination of V c , Qr , X 2 free and R sas metrics have been proposed as more robust evaluation metrics 75 .”

“Evaluation of protein–protein interactions, including oligomerization, is another major potential application of AF2 that continues to be explored 17 .”

“Some groups have used MD simulations to suggest that pLDDT and PAE metrics provide information on dynamics and disorder 42,65 .”

“Initial efforts to enhance sampling of different conformations have involved altering the number of sequences used to generate shallower MSA representations, masking coevolution-ary information provided by MSAs and splitting conflicting coevolutionary signals by clustering MSAs 46,47,105 .”

“New algorithms, such as AlphaFill, are actively being developed that could improve AF2 structure prediction and refinement for cofactor, prosthetic group-or ligand-bound proteins 33 .”

“Native conformations of proteins are often described as time-averaged ensembles of conformations with Boltzmann-type distributions, especially IDPs and proteins with IDRs 100,101 .”

“Terwilliger et al. 61 estimate that ~10% of residues predicted by AF2 with high confidence deviate from the backbone by more than 2 Å from native conformations observed in experimental structures.”

“Yin et al. 55,54 benchmarked 152 heterodimeric protein complexes, revealing that AF2 and AlphaFold2-Multimer had a 51% success rate.”

“In support of this, studies have reported an inability of AF2 to predict the effects of mutations on protein structure and stability 14,49,50 , which may be due to a training bias on stable structures or an inability to extract signal from small mutations through MSAs.”

“In support of this, studies have reported an inability of AF2 to predict the effects of mutations on protein structure and stability 14,49,50 , which may be due to a training bias on stable structures or an inability to extract signal from small mutations through MSAs.”

“Furthermore, the accuracy of PAE graphs for interdomain prediction has not been as extensively benchmarked as for intradomain contacts 36 .”

“When more than one input sequence is provided, AlphaFold-Multimer or AF2Complex is used 20,21 .”

“Some groups have used MD simulations to suggest that pLDDT and PAE metrics provide information on dynamics and disorder 42,65 .”

“…use of AlphaFold models in molecular replacement can be further enhanced by downweighting or removing low-confidence regions 23 . https://doi.org/10.1038/s41589-024-01638-w Another burgeoning area where AlphaFold models are integrated with experimental data is solution-state NMR 15,27,28,30,81–85 .”

“Apart from IDPs, well-folded globular proteins, such as G-protein-coupled receptors and kinases, also sample a wide range of conformations to carry out their biological function 102,103 .”

“Updated and refined approaches for error categorization may provide better methods for model quality assessment relative to pLDDT and PAE metrics 14,67 .”

“The standard implementation of AlphaFold performs well only in detecting a single structural snapshot (the ‘ground state’ structure), likely due to the lack of a large set of redundant protein conformers in the training set 16,32 .”

“The use of AlphaFold models in structure-determination protocols has been shown to reduce the time and effort required relative to ab initio model building 68–70 .”

“These deviations can be broadly categorized into cases with (1) inaccurate secondary, tertiary and/or quaternary the accuracy of the standard implementation of AF2 against different types of nonglobular molecular targets, which could limit its potential applications 15,16 .”

“2b), potentially due to the https://doi.org/10.1038/s41589-024-01638-w inability of AF2 to orient cysteine pairs for disulfide bond formation 31 .”

“Thus, several groups have worked to extend AlphaFold to include predictions of structural ensembles and excited state structures 102,104 .”

“Further benchmarking of modified AF2 protocols against IDPs, IDR-containing proteins and alternative conformation is required to establish protein prediction strengths and limitations for those of systems 43,48 .”

“Another approach is to enable dropout layers in the neural network, which are usually commonly used only during neural network training 8,106,107 .”

“Future benchmarks should continue to evaluate AF2 against panels of IDPs and IDR-containing proteins using novel critical assessment of protein intrinsic disorder targets 44 .”

“McDonald et al. 25 performed a benchmark of 588 peptides, revealing that AF2 predicts many α-helical and β-hairpin peptide structures with surprising accuracy.”

“Yin et al. 55,54 benchmarked 152 heterodimeric protein complexes, revealing that AF2 and AlphaFold2-Multimer had a 51% success rate.”

“Recently, an AF2-integrated iterative procedure for molecular replacement has been developed in which AlphaFold models are used during the initial structure-solution cycle, followed by data-guided cycles of AlphaFold structure prediction and model rebuilding 60,69 .”

“Benchmarking AF2 against membrane proteins represents a challenge, as the membrane environment, which includes lipids and other proteins, is not directly considered by current versions of AlphaFold 38 .”

“Several groups have reported cases of AF2 models with low PAE values ( 5 Å) that deviated from experimental data 62,63 .”

“An important caveat is that one must be wary of overfitting AF2 models to SAXS envelopes, especially for lower-resolution data 74 .”

“An inverted version of AlphaFold, called AlphaDesign, has been used for de novo protein design with some success 53 .”

“Several groups have proposed that AF2 has only learned how to estimate 3D structures using patterns extracted from known folds in the Protein Data Bank (PDB) and coevolutionary information between residues rather than the underlying physical and chemical basis of protein folding 3,10–12 .”

“The most dramatic case is when AF2 provides excellent evaluation metrics despite complete disagreement of the model’s backbone structure with an experimental structure.”

“Refinement of AF2-derived models with NMR-derived restraints, such as chemical shift values, torsion angles, residual dipolar couplings (RDCs) and nuclear Overhauser effect (NOE) data, could help improve accuracy of peptide modeling 15 .”

“However, care must be taken when interpreting predictions, as highlighted by the incorrect AF2 model of the β-endorphin amyloid fiber relative to its experimental solid-state NMR structure 58 (Fig.”

“Preliminary studies suggest that the majority of targets with very low confidence score (pLDDT 50) assigned by AF2 are likely to be IDPs or IDRs rather than well-folded structures that AF2 fails to predict 42–44 .”

“Several groups have proposed that AF2 has only learned how to estimate 3D structures using patterns extracted from known folds in the Protein Data Bank (PDB) and coevolutionary information between residues rather than the underlying physical and chemical basis of protein folding 3,10–12 .”

“Previous analysis suggests that AF2 predicts on average ~50% of residues across all proteins with high confidence 17,59 .”

“Thus, several groups have worked to extend AlphaFold to include predictions of structural ensembles and excited state structures 102,104 .”

“…use of AlphaFold models in molecular replacement can be further enhanced by downweighting or removing low-confidence regions 23 . https://doi.org/10.1038/s41589-024-01638-w Another burgeoning area where AlphaFold models are integrated with experimental data is solution-state NMR 15,27,28,30,81–85 .”

“New methods are fine-tuning AF2-based pipelines for specific types of peptide–protein complexes (that is, peptide–major histocompatibility complex) 26 .”

“…with its deep learning algorithm that enables accurate prediction of three-dimensional (3D) protein structures from only the target amino acid sequence, potentially solving the half-a-century-old protein structure-prediction problem: how to predict 3D structures from only sequence information 1–4 .”

“Recent methods have shown that fitting of SAXS data substantially improves when an ensemble of AlphaFold-predicted structures is used rather than a static AlphaFold model 71 , highlighting the importance of integrating AlphaFold models with experimental data.”

“The use of experimental data, especially SAXS, NMR and cryo-EM, has also been described to guide modeling of ensembles and alternative conformations 71,82,108 .”

“A series of AF2 models can be compared with experimental NMR data in the form of distance-and conformation-sensitive structural restraints obtained from NOEs 81 , RDCs 29,86 , paramagnetic relaxation enhancements 87,88 and/or pseudocontact shifts 89 (Fig.”

“AlphaFold models have also been used extensively together with single-particle cryo-EM data 60,76,95,96 (Fig.”

“…use of AlphaFold models in molecular replacement can be further enhanced by downweighting or removing low-confidence regions 23 . https://doi.org/10.1038/s41589-024-01638-w Another burgeoning area where AlphaFold models are integrated with experimental data is solution-state NMR 15,27,28,30,81–85 .”

“In agreement with these findings, Azzaz et al. 40 have shown the difficulty of AF2 in modeling membrane proteins owing to ‘epigenetic’ factors (that is, lipid environment, co-receptor-induced structural changes, post-translational modifications) that control protein structure beyond the amino acid sequence.”

“The use of experimental data, especially SAXS, NMR and cryo-EM, has also been described to guide modeling of ensembles and alternative conformations 71,82,108 .”

“Refinement of AF2-derived models with NMR-derived restraints, such as chemical shift values, torsion angles, residual dipolar couplings (RDCs) and nuclear Overhauser effect (NOE) data, could help improve accuracy of peptide modeling 15 .”

“Here, NMR assignments for methyl side chain groups can be obtained using only methyl–methyl NOEs obtained from 3D NMR experiments and the atomic coordinates of a structure (or AF2-predicted structure) as input with software such as MAUS, MAGIC and methylFLYA 91,93,94 .”

“However, other reports have compared pLDDT scores with crystallographic B factors to suggest that AF2 confidence metrics are unable to provide direct information on local flexibility 66 .”

“Preliminary reports also suggest that AF2 may also be able to predict oligomeric states of proteins and amyloids 17,57 .”

“…are available (that is, in vitro recombinant protein production or in situ characterization is not possible), insights into the structure and function of proteins may be primarily guided by AF2 predictions supplemented with molecular dynamics simulations to further evaluate model stability 64,65 .”

“A series of AF2 models can be compared with experimental NMR data in the form of distance-and conformation-sensitive structural restraints obtained from NOEs 81 , RDCs 29,86 , paramagnetic relaxation enhancements 87,88 and/or pseudocontact shifts 89 (Fig.”

“Moreover, in the absence of NMR resonance assignments, AF2 models can be used as structural templates toward automated assignment 90 .”

“The use of experimental data, especially SAXS, NMR and cryo-EM, has also been described to guide modeling of ensembles and alternative conformations 71,82,108 .”

“Several groups have reported cases of AF2 models with low PAE values ( 5 Å) that deviated from experimental data 62,63 .”

“Updated and refined approaches for error categorization may provide better methods for model quality assessment relative to pLDDT and PAE metrics 14,67 .”

“Second, AF2 performs poorly for targets embedded in membrane thickness outside the range of 15–35 Å as well as targets with novel features not commonly present in the PDB.”

“5,37).”

“The use of experimental data, especially SAXS, NMR and cryo-EM, has also been described to guide modeling of ensembles and alternative conformations 71,82,108 .”

“A series of AF2 models are fitted into the 3D cryo-EM density maps, and each is evaluated for goodness of fit 96 and can be refined to generate a final structure 97 .”

“The use of experimental data, especially SAXS, NMR and cryo-EM, has also been described to guide modeling of ensembles and alternative conformations 71,82,108 .”

“A series of AF2 models can be compared with experimental NMR data in the form of distance-and conformation-sensitive structural restraints obtained from NOEs 81 , RDCs 29,86 , paramagnetic relaxation enhancements 87,88 and/or pseudocontact shifts 89 (Fig.”

“Preliminary comparison of SAXS-derived versus AlphaFold calculated P ( r ) curves revealed that, for many cases, a static AF2 model does not adequately describe solution-state structures 42,72 .”

“…use of AlphaFold models in molecular replacement can be further enhanced by downweighting or removing low-confidence regions 23 . https://doi.org/10.1038/s41589-024-01638-w Another burgeoning area where AlphaFold models are integrated with experimental data is solution-state NMR 15,27,28,30,81–85 .”

“Here, NMR assignments for methyl side chain groups can be obtained using only methyl–methyl NOEs obtained from 3D NMR experiments and the atomic coordinates of a structure (or AF2-predicted structure) as input with software such as MAUS, MAGIC and methylFLYA 91,93,94 .”

“The use of AlphaFold models in structure-determination protocols has been shown to reduce the time and effort required relative to ab initio model building 68–70 .”

“Furthermore, membrane proteins represent less than 3% of total structures in the PDB 15 , meaning that the training set used by AF2 was highly biased toward soluble proteins 39 .”

“The use of experimental data, especially SAXS, NMR and cryo-EM, has also been described to guide modeling of ensembles and alternative conformations 71,82,108 .”

“Reports suggest that AF2 models should not be fit into cryo-EM maps with resolution greater than 6 Å 98,99 .”

“I believe the revised manuscript is now well-positioned for publication.”

“By using pLDDT scores to identify regions of low confidence, these regions can be targeted in MD simulations to explore potential alternative conformations or to better understand the dynamic behavior of flexible regions [2,8].”

“…regions display incorrect secondary, tertiary, or quaternary structures; (2) examples of backbone structure side chain rotamer or individual domain structures are accurate but misaligned relative to each other; and (3) instances where even high-confidence regions are still inaccurate [37].”