A fully funded PhD studentship in the area of machine learning protein conformational spaces is currently available. The student will work on developing molearn, our generative neural network trainable with protein conformations generated via molecular dynamics simualtions or experiments.
To apply, please follow this link. For informal enquiries, please contact Matteo (matteo.t.degiacomi[AT]durham.ac.uk).
Our work, in collaboration with the Zuber and Posthaus groups in the University of Bern, was just published in EMBO Reports (well done Julia Bruggisser and Ioan Iacovache for the beautiful cover image!).
We investigated the structure, dynamics, and properties beta-toxin, a protein part of the offensive arsenal of Clostridium perfringes bacteria. This protein assembles according to a unique architecture, featuring beta-barrels at both ends. Calculations from Samuel Musson on the newly discovered atomic structure reveal that the toxin should be marginally selective to cations, and that the solvent-exposed barrel is remarkably flexible.
We conclude that beta-toxin could be suitable candidate for small molecule sensing or selective molecule delivery and transport.
Congratulations to…
Welcome to…
This will be a busy month for the group… let us know if you want to meet up!
Matteo, Sam and Cameron will attend the CCPBioSim conference in Edinburgh (6-8 June). The conference will be followed by a MD+ML workshop, where Matteo will show how to use the MDAnalysis and sklearn Python packages to characterize molecular dynamics simulations.
Matteo will then travel to Erice for the MolSim22 conference (25-29 June) and then, along with Sam, to Barcelona for the MMSML Workshop (14-16 July).
Finally, Josh will attend the CCP5 Summer School in Durham (17-28 July).
A fully funded studentship in the area of machine learning for computational biophysics is available in our group, starting October 2022.
All living organisms contain millions of proteins; biopolymers that fold into three-dimensional biologically active structures playing a vital role in the regulation of life and diseases. Research has seen a lot of focus on determining the atomic structure of different proteins. However, the flexible movement of these biopolymers plays a crucial role in their biological (mal)function.
In recent years, machine learning has been revolutionizing the way we interpret data in many scientific areas. For example, the deep neural network AlphaFold2 can predict the 3-dimensional structures of proteins, whose shape is not known experimentally. In our research, we have designed a deep neural network that can also learn an ensemble of structures of specific proteins from molecular simulations. This project builds upon this breakthrough.
In collaboration with the Willcocks group (Department of Computer Science), the student will develop a general neural network capable of learning and predicting the dynamics of any protein. The neural network will be trained with existing and new data you will produce from molecular dynamics simulations. Applications of this work are vast, ranging from understanding the effect of genetic mutations in cancers to informing the design of proteins to carry out a desired function.
Do you have a background in physics, computer science, chemistry, biology, or related discipline, and are keen to develop your computational skills to address biomolecular problems? Then please direct any informal enquiries to Matteo (matteo.t.degiacomi@durham.ac.uk). More information on the application process can be found here. Applications will be considered until the 3rd April.
Are you thinking about writing a Python program to load a bunch of PDB files and measure some of their properties (CCS, interatomic distances, solvent-accessible paths, SAXS, …)? Perhaps you would like to move a bunch of structures around to form a complex in your integrative modelling pipeline, where every subunit may take several conformations? Or how about creating bead models of electron density maps, or the opposite? Do you find it fiddly? We did too, this is why we have created Biobox!
Biobox is a Python package that caters for virtually all our biomolecular modelling needs. Basically every publication we ever produced in the last 8 years did, in a way or another, feature Biobox. Now we are making it public, hopefully you will find as helpful for your work as it was for ours! We have put information about it here, including API and a Jupyter notebook demonstrating its main features.
Congratulations to….
Welcome to…
Proteins shape is constantly changing in response to a variety of biological processes. For example, a protein may be folded into a compact shape and then open up to grab a molecule such as a drug. Predicting what change occurs when you observe proteins behave in an experiment can be difficult. This is because intermediates between two states are by their nature transient, and therefore difficult to observe.
In collaboration with Chris Willcocks in Durham University, We have trained a neural network on protein shapes generated by molecular dynamics simulations, and then asked it to predict possible protein intermediate conformations. We demonstrated that the system can generate conformations that are both physically plausible and close enough to known intermediates. The convolutional architecture of our neural network can accommodate any protein, without the need of being customised. One added advantage that comes with this architecture is transfer learning, which can assist with training on a sparse dataset by first exposing the neural network to a related richer dataset.
This blog post was written with the help of a neural network, that also decided on the title.
JabberDock, our protein-protein docking algorithm, has been extended to handle integral membrane proteins. We have created a benchmark set featuring 20 diverse dimers, and used it to demonstrate that our method yields an acceptable solution in its top 10 candidates in 75% of cases.
The first application of JabberDock for membrane proteins has been in the scope of a collaboration with the Kukura group at the University of Oxford. Our models of the protein bo3 oxidase have contributed demonstrating that mass photometry, a novel label-free single molecule technique to measure the mass of biomolecules, can interrogate membrane proteins.
Protein docking essentially involves taking the atomic structure of two proteins, generating a collection of possible molecular arrangements, and scoring their quality to determine the best one. However, the vast majority of arrangements will feature proteins that either overlap of do not touch. As these cases are uninteresting, generating them and assessing their quality is a waste of time.
Enter shape tracing. In collaboration with the Chris Willcocks group in Durham’s Computer Science Department (in particular, shout-out to Adam Leach), we combine our protein volumetric representation (STID maps) with a generalisation of ray racing capable of detecting the collision of arbitrary three-dimensional shapes.
This enables us to very quickly generate arrangements of proteins that are in contact but do not overlap, speeding up the search for the most suitable model.
Degiacomi research group 