Phil Hinchliffe and Steve Burston, Delhi Kalwan, Jennifer de Jong, Fabio Parmeggiani and Paul Race

This project was made possible thanks to a collaboration between @bristoluni.bsky.social and @ebi.embl.org. A great thanks goes to @robbarringer.bsky.social, Ioannis Riziotis (Crick), Antonina Andreeva,
@alexbateman1.bsky.social and ...

🙏 Special thanks to my amazing collaborators: Ioannis Riziotis, @robbarringer.bsky.social, Antonina Andreeva and to my supervisor @alexbateman1.bsky.social for their invaluable contributions to this work!

💻 Available as a Python package for easy integration into bioinformatics workflows, and accessible via Google Colab for everyone: colab.research.google.com/github/Franc...

🧬 Isopeptor enables reliable detection and geometry evaluation of these covalent links using a template-based strategy powered by pyJess.

🔍 The tool demonstrates a precision of 1.0 and recall of 0.947 in identifying incorrectly modelled isopeptide bonds in PDB structures.

Thanks Gonzalo!

A special thanks goes to my supervisor @alexbateman1.bsky.social
and to @matthiasblum.bsky.social
for their support and guidance. 7/7

Our findings have important implications for improving structure predictions, especially for proteins from organisms with limited representation in sequence databases or for rapidly evolving taxa. 6/7

We show that using high plDDT models as templates can increase the speed of AlphaFold2 as implemented in ColabFold, potentially reducing computational costs and carbon footprint. 5/7

We introduced a novel "Best Pick" strategy that combines predictions made with and without multiple sequence alignment (MSA) information, selecting the model with the highest average plDDT. 4/7

This observation led us to explore whether low-confidence predictions could be improved using high-confidence templates from the same protein family. About one-third of low-confidence structures can be "rescued" to reasonable confidence levels using this method. 3/7

By observing the plDDT distribution within protein domain families, we noticed a certain degree of heterogeneity in the confidence of AlphaFold2 predictions. 2/7