TM-score

Summary

TM-score is an alignment-dependent protein structure similarity term introduced by (1) that is widely used for assessing protein structure prediction methods. It is defined as:

TM - score = ma x \frac{1}{L _{t a r g e t}} i \sum L_{co mm o n} \frac{1}{1 + ( \frac{d _{i}}{d _{0} ( L _{t a r g e t} )} ) ^{2}}

$L_{t a r g e t}$ : length of the amino acid sequence of the target protein $L_{co mm o n}$ : number of residues in both the target and query proteins $d$ : Distance between pairs of residues $d_{0} (L_{t a r g e t}) = 1.24 * 3 L_{t a r g e t} - 15 - 1.8$ : Distance scaling factor

Zhang Y, Skolnick J. Scoring function for automated assessment of protein structure template quality. Proteins: Structure, Function, and Bioinformatics. 2004;57(4):702–10. Available from: https://doi.org/10.1002/prot.20264

Quartz 4

Explorer

TM-score

Summary

AlphaFold3 ipTM can distinguish between antibody binders and nonbinders

Protein structure prediction and design confidence metrics do not correlate with binding affinity

Diffusion-based structure prediction can sometimes model conformational ensembles

Heavy and light chains of de novo designed antibodies with the same framework and binding mode can be exchanged

Including structure prediction confidence while training inverse folding improves sequence diversity but not sequence recovery

Inverse folding sequence perplexities correlate with Rosetta energies, forward folding TM-scores, and sequence recovery

Natural sequences have higher pTM but lower pLDDT than de novo sequences

PLM-based sequence searches outperform sequence- and matches structure-based search methods

Protein backbones designed using diffusion, but not sequence-based models, have fewer beta sheets

Protein property prediction using PLMs does not benefit from scale except when predicting inferring features of either structural or sparsely populated sequence families

Reinforcement learning outperforms fine-tuning in property-guided inverse folding

Sequence perplexity and TM-score are negatively correlated when predicting structure using protein language models

Structure prediction from conformational samplers tends to be worse than those from vanilla structure prediction methods

Structure prediction metrics can be used to predict oligomer stoichiometry

Structure prediction uncertainty metrics as energy functions

Template CDR retrieval using embeddings

Unfreezing PLM during structure training improves prediction quality

pTM is less lenient than pLDDT toward structural novelty when predicting alternate conformations

Confidence metrics for diffusion-based structure prediction methods can be improved with minimal changes to conditioning representations

Confidence metrics from structure prediction correlate with accuracy even in the absence of coevolutionary data