Predicting protein conformational flexibility remains a major challenge in structural biology. While we can now accurately model static protein structures, understanding their dynamics is still difficult, largely due to a lack of suitable training data.
New OpenFold3 preview out! (OF3p2)
It closes the gap to AlphaFold3 for most modalities.
Most critically, we're releasing everything, including training sets & configs, making OF3p2 the only current AF3-based model that is functionally trainable & reproducible from scratch🧵1/9
I’m excited to share my first peer-reviewed publication, and my first first-author paper, "Time in mind: a multidisciplinary review on temporal perception, cognition, and memory" is now published open access in Frontiers in Cognition!
www.frontiersin.org/journals/cog... #psychology #science #time
SAbDab2 collects, annotates, and organises all antibody structures in the PDB.
This includes paired- and single-chain antibodies (like VHHs) in various formats, and antibody–antigen complexes.
New data is added weekly.
Many thanks to my collaborators Henriette Capel, Ben Williams, and Chris Taylor; to our incredible PI Charlotte Deane; and everyone else at the Oxford Protein Informatics Group @opig.stats.ox.ac.uk !
With improvements to our annotation pipeline, SAbDab2 contains more structures than ever before.
With this release, we are introducing support for VNARs and a variety of novel antibody formats.
Today, we're announcing SAbDab2!
In brief:
- clean, pre-processed antibody structure data for ML with standardised train/test splits
- massive improvements in structure organisation and annotation consistency
- support for VNARs and antibody construct annotation
- more structures than ever before
Currently, published antibody structures are highly sequence-redundant.
In SAbDab2, every antibody is given a stable SAbDab2 ID, which attaches to every known structure of that antibody across the entire PDB. This makes it easy to compare different conformations, bound states and more.
We look forward to seeing what you do with SAbDab2!
Check it out: sabdab2.opig.stats.ox.ac.uk
We are also releasing antibody and AbAg complex structures processed specifically for AI/ML, alongside ready-made, similarity-based train/test splits.
Unlike commonly used date-splits, these should mitigate against test-set contamination with structures similar to those seen in training.
