Ora Furman Lab 馃帡 (@FurmanLab) header
Ora Furman Lab 馃帡 (@FurmanLab) profile picture

Structural basis of peptide-mediated protein communication and function @HebrewU Member of @RosettaCommons. Tweets by lab members. 馃 https://teapawt.org/t.co/UMEreURWke 馃

Jerusalem, Israel
Joined August 2020
5/ This was an amazing collaboration with the @lab_wiener and @RouvinskiLab - thank you! 馃檶 Graphical abstract below 猬囷笍 #SARSCoV2 #COVID19 #Virology #CellReports #ComplementSystem #Immunology
4
122
4/ 馃挕 Heparin analogs (like Enoxaparin) block NP surface binding, preventing immune complex formation and rescuing cells from complement attack. A possible path to reduce immunopathology.
1
1
111
2/ NP transfers from infected to neighboring uninfected cells, where it binds heparan sulfate proteoglycans (HSPGs) via its positively charged RNA-binding regions. This creates targets for anti-NP IgGs to trigger complement activation.
1
1
21
Here are two examples: in the first, ipTMdrops upon extension, even though the motif is confidently modeled. actifpTM stays in the same range. In the 2nd example, extension 2 takes a different conformation than the solved structure, which the actifpTM value correctly follows.
1
3
502
In the output, we provide actifpTM for all pairs of chains, as well as a general value for the whole complex. To allow for comparison, a pairwise ipTM, as well as pTM-score for each chain separately also get calculated.
1
3
490
We call the new metric actifptm (actual interface pTM). We implemented its calculation in ColabFold (also localcolabfold), and the calculation can be invoked with checking the calc_extra_ptm box on Colab, or adding the flag 鈥揷alc-extra-ptm.
1
4
521
By using the 64 predicted error bins, a pairwise pTM-score matrix is calculated, then multiplied by the mask. We amended this mask by using the predicted distogram, with a cutoff of 8A between Cb atoms. Now the calculation should only focus on actual interface residues.
1
4
517
In most scientific discussion, ipTM in AF2 has been referred to as interface pTM. However, its calculation does not happen on the interfaces as we usually define them, but by taking into account all inter*chain* residue pairs with equal weights (normalized interchain mask below).
1
2
617
ipTM is calculated from 64 predicted error bins. When comparing those, adding more flanking regions around the binding regions usually does not change the matrix. So, if we only calculate over the actual interface residues, we should get results that are similar to each other.
1
4
696
Our work on refining AlphaFold2鈥檚 ipTM has been accepted for publication in Bioinformatics @OxfordJournals We have been questioning why ipTM worsens when docking peptides with disordered flanking regions, and modified the calculation to remove this effect. Let鈥檚 dive in! 馃憞
2
35
6
164
25,770
With lots of help from Sergey Liskov, the FlexPepDock server has been moved to the ROSIE2 server. r2.graylab.jhu.edu/apps/ Check out and let us know if there are any issues!
7
29
1,460
8FG6 with AF2 - a clear cut case, predicted perfectly with either AF models.
1
195
7UE2 with AF2 - again, the repeats are bound to mess up the predictions. A confident peptide can confuse the places of P and L but models with the correct side-chains in the correct places are not necessarily confident...
1
139
To be fair, we also predicted this with only the solved region. This way, the peptide was in the correct pocket, but shifted or flipped. At least the confidence now correctly showed that this is not the right position.
94
8HEP with AF2 - also spot on. AlphaFolds (also other modeling suits) really like beta-sheet complementation. Unfortunately, they also tend to be overfavored...
1
155
8SG2 with AF2 - ran without phosphorylation, which obviously does not help. Unfortunately, it is confident in part of the interaction that is absolutely inaccurate :(
68
7X5C with AF2 - only showing one of the models, but all are messed up. In the top ranked, the loop became a helix on the other side of the protein 馃槷 - the peptide is also misplaced as a result.
74
7DEE with AF2 - still poorly predicted, but at least the confidence metrics (plddt < 50 and ipTM <0.55) reflects this.
88
7AAM - an extended peptide binding a helical bundle - this prediction is confidently spot on!
1
1
8
2,473
7PIM - the binding site is found, however the peptide is badly modeled. At least the confidence scores reflect this.
1
2
2,143