This model uses #AlphaFold2-like architecture to perform flexible docking of small molecules into #protein targets, predicting both structure and affinity. Read the paper: https://t.co/9sbF5ZeHC3

"#CombFold: predicting structures of large protein assemblies using a combinatorial assembly algorithm and #AlphaFold" is now online in @naturemethods! @DinaSchneidman Read the paper:

8/8 CombFold is available from https://t.co/Du0SBLFHsB and also as a Colab notebook. Please send us your feedback if you try it.

7/8 Unknown stoichiometry remains one of the most significant gaps in high-throughput complex prediction. We present a proof-of-concept of using CombFold for rapid enumeration of stoichiometries.

6/8 Many experimentally solved structures of large complexes have unresolved regions or even domains. On our benchmark, CombFold extended the structural coverage with high confidence by 20% on average. Here, the newly predicted domains are shown as cartoons.

5/8 We used CombFold to model the unknown structure of the human Elongator holoenzyme complex. The model has high confidence and is similar to partial homologous structures.

4/8 We have benchmarked CombFold on 60 large asymmetric complexes with 1,300 to 18,000 amino acids and 5 to 30 chains. CombFold successfully assembled 70% of the complexes among Top-10 predictions with TM-score > 0.7.

3/8 CombFold is a combinatorial and hierarchical assembly algorithm. Starting from individual subunits it joins pairs of substructures iteratively to construct the complex. Multiple assembly possibilities are explored through the enumeration of assembly trees.

2/8 Large complexes are hard to predict accurately even with #AlphaFold-multimer. Turns out, you can predict structures for pairs of subunits separately using #AlphaFold and then assemble them into a complex!

1/8 Excited to introduce CombFold - an algorithm for predicting large multi-protein complexes using the combinatorial and hierarchical assembly of #AlphaFold models. @DinaSchneidman https://t.co/sW5gMgf8Ky