Research Group Leader Senior
- Structural biology of protein-DNA complexes
- Phosphorylation mechanisms in Wnt signaling
- Small molecule inhibitors
- NMR methods
- Biochemical and structural characterization of protein-DNA interactions involved in DNA repair and transcription
- Design and pre-clinical evaluation of compounds for lead optimization
- Transforming biomolecular NMR to stay at the forefront of Structural Biology
Content of research
Ultrafast methods for NMR structure determination
The three-dimensional structure of a protein provides crucial insights into its biological function. NMR is the method of choice for proteins that are difficult, if not impossible, to crystallize or are relatively small (<50 kDa) for cryogenic electron microscopy to obtain atomic resolution. Yet, NMR data analysis is a laborious task and the road to NMR structure is generally a long one. The chief barrier is that existing methodologies require a significant amount of spectrometer time (several weeks), and effort by a trained expert (up to several months). We combine 4D NMR Spectroscopy with artificial intelligence to meet the key objectives of NMR structure determination; minimal data collection, least human intervention, and applicability to large proteins. We have developed 4D-CHAINS technology that utilizes only two 4D spectra and offers several advantages: (i) increased predictive power and higher reliability in sequential mapping by using all available correlated aliphatic chemical shifts, (ii) fully automated assignments of NMR chemical shifts at >95% completeness with <1.5% error rate, (iii) tremendous reduction in human effort and NMR spectrometer time needed to obtain data-driven, high-resolution structures, as the NOESY spectrum contains a set of distance constraints for structure calculations.
Mechanisms of protein phosphorylation
Phosphorylation, mediated by protein kinases, is a common posttranslational modification with 500,000 potential phosphorylation sites in the human proteome and 25,000 phosphorylation events described for 7,000 human proteins. Casein kinases are critical regulators of multiple developmental signaling pathways such as Wnt/β-catenin, Wnt/planar cell polarity, or Hedgehog. We use a suite of biochemical and structural biology tools in order to gain a mechanistic view of CK1ε action in the Wnt signaling pathways. Dishevelled (DVL) is a key target of CK1ε in the Wnt pathways, yet the mechanistic basis of DVL phosphorylation-driven activation is not understood. We apply integrated structural biology to (i) probe the DVL conformational landscape using in vitro and in vivo FRET sensors coupled to SAXS and CryoEM, (ii) understand the (auto)phosphorylation regulatory mechanisms of CK1ε, (iii) analyse by NMR the functional consequences of DVL phosphorylation and (iv) monitor DVL phosphorylation by real-time NMR under controlled cellular conditions.
Receptor-ligand interactions have everlastingly been the center of research in pharmaceutical sciences. A common trend in drugging cancer and other disease targets is to optimize the ligand structure in order to increase the binding affinity and specificity to the receptor. We develop deepScaffOpt and deepHitMiner, two algorithms that employ machine learning for predicting the ligand binding affinity to a receptor. deepScaffOpt is a Regressor (predicts a precise free energy of binding) while deepHitMiner is a Classifier (returns a probability of a small molecule being a binder or a non-binder). deepScaffOpt is used when one needs to rank a small number of molecules (<1000) according to their predicted binding affinity, while deepHitMiner is suitable for screening chemical libraries of millions of small molecules to identify new hits (molecules that bind to the receptor protein), but also for off-target identification (one molecule binds to multiple protein receptors) and drug repurposing (an approved drug that binds to multiple target proteins can be repurposed to treat another disease). In both algorithms small molecules are described in a vector form, which permits the use of machine learning. As a proof-of-concept, deepScaffOpt outperformed most of the other algorithms in the D3R Grand Challenge 2017.