Why AlphaFold is Not Like AlphaGo

2021 ◽  
Vol 2021 (02) ◽  
pp. 0206
Author(s):  
Terry Bollinger
Keyword(s):  
Protein Structure ◽  
Prediction Accuracy ◽  
Human Learning ◽  
Game Playing ◽  
Human Knowledge ◽  
Experimental Time ◽  
Learning Speed ◽  
Protein Prediction

AlphaFold2 is the second major iteration of a protein structure predictor by Google-owned DeepMind Lab. DeepMind is famous for creating AlphaGo Zero, the first game-playing system to transcend the rules taught by human trainers. When AlphaFold2 made a significant leap in protein prediction accuracy in the fourteenth annual CASP competition, even reserved publications like Nature were noticeably breathless in their praise of the results. It was not just the impressive and well-proven leap in prediction accuracy that made AlphaFold2 notable, but also its association with the DeepMind brand and implicitly with the beyond-human learning successes of AlphaGo Zero. But is this latter component of its notoriety and acclaim justified? That is, beyond superficial name similarities, is the design of AlphaFold2 sufficiently like that of AlphaGo Zero to enable a similar leap ahead of human knowledge and expertise? An analysis of the underlying designs says no. In contrast to the fully virtualized, faster-than-human learning speeds of AlphaGo Zero, the learning speed of AlphaFold2 remains firmly attached to and limited by human experimental time. AlphFold2 thus is inherently incapable of the trans-human leaps in learning speed demonstrated by AlphaGo Zero.

A New Prediction Method for Chaotic Time Series

2013 ◽  
Vol 380-384 ◽  
pp. 1673-1676
Author(s):  
Juan Du
Keyword(s):  
Neural Network ◽  
Time Series ◽  
Prediction Accuracy ◽  
Time Series Prediction ◽  
Prediction Method ◽  
Cumulative Effect ◽  
Training Data ◽  
Training Algorithm ◽  
Learning Speed ◽  
Sunspot Data

In order to show the time cumulative effect in the process for the time series prediction, the process neural network is taken. The training algorithm of modified particle swarm is used to the model for the learning speed. The training data is sunspot data from 1700 to 2007. Simulation result shows that the prediction model and algorithm has faster training speed and prediction accuracy than the artificial neural network.

A Constant Proportion in Protein Structure

2018 ◽  
Author(s):  
Francisco Javier Lobo-Cabrera
Keyword(s):  
Protein Structure ◽  
Spatial Clustering ◽  
A Priori ◽  
Data Bank ◽  
X Ray ◽  
Protein Size ◽  
Protein Prediction ◽  
Quality Check ◽  
First Time ◽  
Secondary Structure Composition

The principles governing protein structure are largely unknown. Here, a structural proportion universal (R2 = 0.978) among proteins is reported. The model variance is shown to be independent from protein size, secondary structure composition, compactness or relative surface area. The structural characteristic under study --named here QUILLO-- quantifies residue-type spatial clustering. In this way, polar, hydrophobic, acidic and basic residues are evaluated individually and their values added up. For the analysis, all X-Ray currently determined structures deposited in the Protein Data Bank were studied. The QUILLO proportion offers for the first time an a priori protein prediction quality-check. Indeed, predictions with unexpected proportion values correspond to low ranks in the CASP12 experiment. The reason behind a specific, constant rule for protein folding remains unknown.

scholarly journals A de novo protein structure prediction by iterative partition sampling, topology adjustment, and residue-level distance deviation optimization

2021 ◽  
Author(s):  
Jun Liu ◽  
Kai-Long Zhao ◽  
Guang-Xing He ◽  
Liu-Jing Wang ◽  
Xiao-Gen Zhou ◽  
...  
Keyword(s):  
Protein Structure ◽  
Dihedral Angle ◽  
Structure Prediction ◽  
Prediction Accuracy ◽  
Closed Loop ◽  
De Novo ◽  
Residue Level ◽  
Fragment Assembly ◽  
Assembly Method ◽  
Distance Deviation

Abstract Motivation With the great progress of deep learning-based inter-residue contact/distance prediction, the discrete space formed by fragment assembly cannot satisfy the distance constraint well. Thus, the optimal solution of the continuous space may not be achieved. Designing an effective closed-loop continuous dihedral angle optimization strategy that complements the discrete fragment assembly is crucial to improve the performance of the distance-assisted fragment assembly method. Results In this article, we proposed a de novo protein structure prediction method called IPTDFold based on closed-loop iterative partition sampling, topology adjustment and residue-level distance deviation optimization. First, local dihedral angle crossover and mutation operators are designed to explore the conformational space extensively and achieve information exchange between the conformations in the population. Then, the dihedral angle rotation model of loop region with partial inter-residue distance constraints is constructed, and the rotation angle satisfying the constraints is obtained by differential evolution algorithm, so as to adjust the spatial position relationship between the secondary structures. Lastly, the residue distance deviation is evaluated according to the difference between the conformation and the predicted distance, and the dihedral angle of the residue is optimized with biased probability. The final model is generated by iterating the above three steps. IPTDFold is tested on 462 benchmark proteins, 24 FM targets of CASP13, and 20 FM targets of CASP14. Results show that IPTDFold is significantly superior to the distance-assisted fragment assembly method Rosetta_D (Rosetta with distance). In particular, the prediction accuracy of IPTDFold does not decrease as the length of the protein increases. When using the same FastRelax protocol, the prediction accuracy of IPTDFold is significantly superior to that of trRosetta without orientation constraints, and is equivalent to that of the full version of trRosetta. Availability The source code and executable are freely available at https://github.com/iobio-zjut/IPTDFold. Supplementary information Supplementary data are available at Bioinformatics online.

scholarly journals A de novo protein structure prediction by iterative partition sampling, topology adjustment, and residue-level distance deviation optimization

2021 ◽  
Author(s):  
Jun Liu ◽  
Kailong Zhao ◽  
Guangxing He ◽  
Liujing Wang ◽  
Xiaogen Zhou ◽  
...  
Keyword(s):  
Protein Structure ◽  
Dihedral Angle ◽  
Structure Prediction ◽  
Prediction Accuracy ◽  
Closed Loop ◽  
De Novo ◽  
Residue Level ◽  
Fragment Assembly ◽  
Assembly Method ◽  
Distance Deviation

Motivation: With the great progress of deep learning-based inter-residue contact/distance prediction, the discrete space formed by fragment assembly cannot satisfy the distance constraint well. Thus, the optimal solution of the continuous space may not be achieved. Designing an effective closed-loop continuous dihedral angle optimization strategy that complements the discrete fragment assembly is crucial to improve the performance of the distance-assisted fragment assembly method. Results: In this article, we proposed a de novo protein structure prediction method called IPTDFold based on closed-loop iterative partition sampling, topology adjustment and residue-level distance deviation optimization. First, local dihedral angle crossover and mutation operators are designed to explore the conformational space extensively and achieve information exchange between the conformations in the population. Then, the dihedral angle rotation model of loop region with partial inter-residue distance constraints is constructed, and the rotation angle satisfying the constraints is obtained by differential evolution algorithm, so as to adjust the spatial position relationship between the secondary structures. Lastly, the residue distance deviation is evaluated according to the difference between the conformation and the predicted distance, and the dihedral angle of the residue is optimized with biased probability. The final model is generated by iterating the above three steps. IPTDFold is tested on 462 benchmark proteins, 24 FM targets of CASP13, and 20 FM targets of CASP14. Results show that IPTDFold is significantly superior to the distance-assisted fragment assembly method Rosetta_D (Rosetta with distance). In particular, the prediction accuracy of IPTDFold does not decrease as the length of the protein increases. When using the same FastRelax protocol, the prediction accuracy of IPTDFold is significantly superior to that of trRosetta without orientation constraints, and is equivalent to that of the full version of trRosetta.

scholarly journals A systematic structural comparison of all solved small proteins (4-6 kDa) reveals the weight of disulfide bonds in proteins' foldability

2021 ◽  
Author(s):  
Mariana Hoyer Moreira ◽  
Fabio C. L. Almeida ◽  
Tatiana Domitrovic ◽  
Fernando L. Palhano
Keyword(s):  
Protein Structure ◽  
Disulfide Bond ◽  
Structure Prediction ◽  
Disulfide Bonds ◽  
Hydrophobic Core ◽  
Structural Comparison ◽  
Small Proteins ◽  
Protein Prediction ◽  
Aqueous Solvent ◽  
Tertiary Protein Structure

Defensins are small proteins, usually ranging from 4 to 6 kDa, amphipathic, disulfide-rich, and with a small or even absent hydrophobic core. Since a hydrophobic core is generally found in globular proteins that fold in an aqueous solvent, the peculiar fold of defensins can challenge tertiary protein structure predictors. We performed a PDB-wide survey of small proteins (4-6 kDa) to understand the similarities of defensins with other small disulfide-rich proteins. We found no differences when we compared defensins with non-defensins regarding the proportion and exposition to the solvent of apolar, polar, and charged residues. Then we divided all small proteins (4-6 kDa) deposited in PDB into two groups, one group with at least one disulfide bond (bonded, defensins included) and another group without any disulfide bond (unbonded). The group of bonded proteins presented apolar residues more exposed to the solvent than the unbonded group. The ab initio algorithm for tertiary protein structure prediction Robetta was more accurate to predict unbonded than bonded proteins. Our work highlights one more layer of complexity for the tertiary protein prediction structure: small disulfide-rich proteins' ability to fold even with a poor hydrophobic core.

Sign in / Sign up

Export Citation Format

Share Document