State-of-the-Art Estimation of Protein Model Accuracy Using AlphaFold

By James P. Roney, Sergey Ovchinnikov. 2022

223 citations recorded by the source corpus.

The problem of predicting a protein’s 3D structure from its primary amino acid sequence is a longstanding challenge in structural biology. Recently, approaches like AlphaFold have achieved remarkable performance on this task by combining deep learning techniques with coevolutionary data from multiple sequence alignments of related protein sequences. The use of coevolutionary information is critical to these models’ accuracy, and without it their predictive performance drops considerably. In living cells, however, the 3D structure of a protein is fully determined by its primary sequence and the biophysical laws that cause it to fold into a low-energy configuration. Thus, it should be possible to predict a protein’s structure from only its primary sequence by learning a highly-accurate biophysical energy function. We provide evidence that AlphaFold has learned such an energy function, and uses coevolution data to solve the global search problem of finding a low-energy conformation. We demonstrate that AlphaFold’s learned potential function can be used to rank the quality of candidate protein structures with state-of-the-art accuracy, without using any coevolution data. Finally, we explore several applications of this potential function, including the prediction of protein structures without MSAs.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

AFsample: improving multimer prediction with AlphaFold using massive sampling (2023, background): “For complex cases, simply increasing the number of sampled models might not be enough if the evolutionary constraints have trapped the prediction in a local minimum in the conformational landscape (Roney and Ovchinnikov 2022) or the if the evolutionary constraints are weak.”
OpenFold: retraining AlphaFold2 yields new insights into its learning mechanisms and capacity for generalization (2023, methodology): “…and to employ it as a module within novel algorithms, including ones for protein complex prediction (Baek 2021), peptide-protein interactions (Tsaban et al. 2022), structure ranking (Roney and Ovchinnikov 2022), and more ( e.g. , Baltzis et al. 2022, Bryant et al. 2022, Wayment-Steele et al. 2022).”
Improved multimer prediction using massive sampling with AlphaFold in CASP15 (2023, background): “…and Wallner, 2022).48 In cases where the evolutionary constraints have trapped the prediction in a local minimum49 in the conformational landscape or the evolutionary constraints are weak, simply increasing the50 number of sampled models may not be sufficient (Roney and Ovchinnikov, 2022).”
Improving peptide-protein docking with AlphaFold-Multimer using forced sampling (2021, background): “…the quality of structures derived from template information without any MSA information at all, it cannot efficiently sample the folding landscape without MSAs, and if given template information with no MSA information AlphaFold will essentially copy the template (Roney and Ovchinnikov, 2022).”
Single-sequence protein structure prediction using language models and deep learning (2022, methodology): “We refined structures predicted by RGN2 using an AF2Rank-based protocol 28 that imputes the backbone and sidechain atoms.”
Benchmarking AlphaFold for protein complex modeling reveals accuracy determinants (2021, methodology): “50 As the AlphaFold-Multimer algorithm generates an interface pTM score (ipTM) which is used in conjunction with pTM to compute model scores, 36 we examined the use of ipTM score alone in model accuracy discrimination for models from the set of cases in Table 2, and ipTM alone showed promising…”