Since AlphaFold2 “won” the CASP competition in 2020, intrinsically disordered regions (IDRs) have been one of the points of discussion when it comes to AlphaFold2 performance. When the method was published, it was noted that the local predicted LDDT (pLDDT) score correlated well with IDRs. In this short communication, the authors explore this concept further and experiment with other metrics, such as relative solvent accessibility, in order to benchmark IDR prediction by AlphaFold2. Additionally, they also evaluate how pLDDT and relative solvent accessibility can be used to identify regions that fold conditionally upon binding. The conclusion is that AlphaFold2 performs well in the prediction of IDRs but is two orders of magnitude slower and, thus, not the best choice for fast searches.
Like missing those couple of seconds giving out clues about who killed Mr. Thrombey, transitory high-energy conformations can contain crucial information about the mechanism of actions of proteins. In this paper, the authors describe a way to access them using paramagnetic NMR experiments, through the use of pseudocontact shifts (PCS). These PCS provide long-distance structural information that can be used as restraints to determine conformations using simulated annealing, starting from known structures. This method was applied to the adenylate kinase (Adk) metalloprotein enzyme for which only the closed and open conformations are known, and confirmed several observations made by FRET or meta-dynamics simulations. They also show that the method can be generalized to non-metalloproteins by using a metal-binding tag.
Structure-informed microbial population genetics elucidate selective pressures that shape protein evolution
A novel use-case for structure prediction, microbial population genetics. The authors look into evolutionary dynamics in a popular ocean surface microbe taxon through the lens of protein structure as predicted by AlphaFold2. For this, they calculate the rate of synonymous and non-synonymous mutations at each single codon variant site and compare this to relative solvent accessibility and distance from the (predicted) ligand site for the amino acid under consideration. Overall, synonymous variants were way more common than non-synonymous, but the rates of non-synonymous mutations displayed some pretty interesting trends. They were a lot more common in exposed sites vs. buried sites, and they tended to avoid active sites and ligand-binding sites, the first likely due to maintaining structural stability and the second to maintain function. For one protein involved in recycling nitrogen, correctly taking into account its quaternary structure brought forth the expected conclusion that non-synonymous variation again avoids buried and active sites in this protein. This, combined with possible errors in the prediction of ligand binding sites and the steric configurations of residues, means that care needs to be taken in such large-scale structural analyses - though the kinds of conclusions one can draw from them are definitely unique and complementary to the common nucleotide sequence-based analysis.
Figure of the Week
Fig. 3: Linkers L1 and L2 mediate the structural transition to the active state.
a, Overview of the 18–20 MM active conformation. b, c, Detailed view of HNH (b) and RuvC (c) active sites. d, Docking of the L1 linker helix against the PAM-distal gRNA–TS duplex, shown as an isosurface representation. e, Interactions of L1 and L2 regions with the minor groove of the gRNA–TS duplex. HNH extending from L1 and L2 linkers has been removed for clarity and does not interact with this region of the gRNA–TS duplex.
- FastFold: Reducing AlphaFold Training Time from 11 Days to 67 Hours
- Jalview with a pretty sweet update
Jalview 184.108.40.206 is out!— Jalview.org (@Jalview) March 10, 2022
This 'minor release' adds
- @UCSFChimeraX, #Pymol (ce/incentive)
- discovery of structures from 3D-Beacons @emblebi including #alphafold and @SWISS_MODEL
- a new command line tool
Dowload & release notes via https://t.co/hMw7J57lAF pic.twitter.com/hXAiXpERLV
- MolEvolvR: A web-app for characterizing proteins using molecular evolution and phylogeny
- What have we learned from design of function in large proteins?
- CoDNaS-Q: a database of conformational diversity of the native state of proteins with quaternary structure
- Structural basis for mismatch surveillance by CRISPR–Cas9
- How does a small molecule bind at a cryptic binding site?
- AtomNet PoseRanker: Enriching Ligand Pose Quality for Dynamic Proteins in Virtual High-Throughput Screens
- A new OpenMM Jupyter notebook
We've added a @ProjectJupyter notebook illustrating how to use openmm-torch to run a GPU-accelerated simulation of alanine dipeptide using the ANI-2x ML potential from @adrian_roitberg @olexandr @gpusciguy within OpenMM!— OpenMM (@openmm_toolkit) March 4, 2022
Check it out on @GooogleColab:https://t.co/IwrCLPXVNF
- Residue Assignment in Crystallographic Protein Electron Density Maps with 3D Convolutional Networks
- The Carbon Footprint of Bioinformatics
- Sampling alternative conformational states of transporters and receptors with AlphaFold2
- Protein-Ligand Interaction Graphs: Learning from Ligand-Shaped 3D Interaction Graphs to Improve Binding Affinity Prediction
- Accelerating therapeutic protein design
- On the Value of Using 3D Shape and Electrostatic Similarities in Deep Generative Methods
- An update of AlphaFold-Multimer that helps with the steric clash issues that were cropping up
We have updated the AlphaFold-Multimer paper and code with a new set of protein complex models trained with an improved loss. These models reduce the number of steric clashes and improve accuracy.— Tim Green (@tfgg2) March 10, 2022
Paper: https://t.co/A3uTYVoTeq pic.twitter.com/SHKZzhAX82