Understanding AlphaFold Metrics in Structure Evaluation

AlphaFold is a deep learning model developed by the DeepMind team at Google for predicting the three-dimensional structure of proteins. In case, you are new to AlphaFold, here is a quick introduction to the program.

AlphaFold uses a neural network to predict the three-dimensional structure of a protein from its amino acid sequence. The neural network is trained on a large dataset of known protein structures and amino acid sequences, using a technique called supervised learning. The model uses a combination of convolutional neural networks (CNNs) and recurrent neural networks (RNNs) to learn the complex relationships between protein sequences and structures. The output of the AlphaFold model is a prediction of the three-dimensional structure of the protein, represented as a set of coordinates for each atom in the protein. These predictions are evaluated using a range of metrics, including the predicted local distance difference test (PLDDT), predicted torsion angle metric (PTM), and inferred PTM (iPTM), among others. 

Understanding Structural Metrics Used by AlphaFold

IDDT, PLDDT, PTM, iPTM, and iTol scores are all measures used to assess the accuracy of protein structure predictions made by AlphaFold.

IDDT: The Inter-Residue Distance Test (IDDT) is a measure of the accuracy of predicted residue-residue contacts in a protein structure. It measures the percentage of predicted contacts that are within a specified distance cutoff of their corresponding contacts in the native structure. The IDDT score ranges from 0 to 100, with higher scores indicating better accuracy.

PLDDT: Predicted Local Distance Difference Test (PLDDT) is the most commonly used metric to evaluate the overall accuracy of AlphaFold protein structure predictions. It is an average of the local distance difference test (lDDT) scores across all residues in the protein. The lDDT score measures the similarity of the predicted and native structures at the local level, and ranges from 0 to 1. The PLDDT score ranges from 0 to 100, with higher scores indicating better accuracy.

PTM: Predicted Torsion Angle Metric (PTM) measures the accuracy of the predicted backbone dihedral angles (phi and psi angles) in the protein structure. It ranges from 0 to 100, with higher scores indicating better accuracy.

iPTM: Inferred PTM (iPTM) is a measure of the accuracy of predicted side-chain conformations. It is calculated based on the agreement between the predicted and observed side-chain rotamer conformations in the native structure, and ranges from 0 to 100.

iTol: Interpolated thermal stability (iTol) is a measure of the predicted stability of a protein structure. It is calculated based on the predicted energy landscape of the protein and provides an estimate of the melting temperature of the protein. Higher iTol scores indicate greater predicted stability.

All of these scores provide different measures of the accuracy of AlphaFold protein structure predictions and can be used in combination with other experimental data and biological knowledge to gain insight into the structure and function of proteins.