Computational Structural Biology Lab · IBG, Izmir
We combine AlphaFold-based sampling, molecular dynamics and integrative modelling to assess the accuracy of predicted assemblies, resolve the interface contacts that determine their binding specificity, and extract conformational dynamics from sequence-derived information such as AlphaFold distograms.
Our research moves from static to dynamic views of biomolecular interfaces: we test how reliably they can be predicted, map how mutations rewire them, resolve which persistent contacts govern binding, and ask how much of this behaviour is encoded in sequence.
How reliable are structure and affinity predictions?
Which mutations rewire an interface?
Which persistent contacts decide partner choice and function?
How much of an interface’s motion is written in its sequence?
feeds back into prediction
AlphaFold-based pipelines model most assemblies well, but not all. Through CASP, CAPRI and dedicated benchmarks, we test how far structure and affinity predictors can be trusted, and build leaner sampling and scoring protocols.
A static interface already shows which positions matter. We map how mutations, disease variants and isoforms strengthen, weaken or abolish interfaces, and use these maps for functional characterisation or to redesign binders.
Many interface contacts are transient, and the ones that persist govern binding. We quantify contact persistence from simulations and compare related complexes to find the dynamic networks behind partner choice and function.
What predictors learn from sequence also reflects flexibility. We extract this signal from AlphaFold distograms and AlphaFold-based sampling to predict interface motions and ensembles directly from sequence, without long simulations.
Open-source tools from the lab, with code released on GitHub.
Try it in a HADDOCK tutorial
PROT-ON is part of the official HADDOCK antibody–antigen tutorials, where it scans the docked interface for mutations that improve binding and helps identify an optimal binder sequence.
Principal Investigator
Group Leader, Izmir Biomedicine and Genome Center · Associate Professor and Head of the Molecular Biology and Genetics Program, Dokuz Eylül University
Ezgi trained in chemical engineering at Boğaziçi University and did her PhD with Alexandre Bonvin at Utrecht University, where she became a long-time contributor to the HADDOCK integrative modelling platform. As a Humboldt postdoctoral fellow at EMBL Heidelberg she worked on integrative structure determination of molecular machines, before starting her group at IBG in 2017 with a Humboldt return fellowship.
She was an assessor of protein assemblies in CASP14 and CASP15, received the Science Academy Young Scientist Award (BAGEP) in 2020, and is the BioExcel Ambassador for Türkiye.
Computations run on MareNostrum 5 at the Barcelona Supercomputing Center, through Türkiye’s EuroHPC partnership, and on in-house GPU workstations.
Three PhD and seven MSc theses supervised as main advisor.
Full texts: search “Ezgi Karaca” as advisor (Danışman) at the National Thesis Centre.
We welcome students and postdocs interested in structure prediction, molecular dynamics and protein interfaces. Send Ezgi a short note on your interests, with a CV.