Modulation of receptor function and enzyme evolution via optimized mutagenesis designs
||Modulation of receptor function and enzyme evolution via optimized mutagenesis designs|
||Hugo Gutierrez de Teran <firstname.lastname@example.org>|
||2022-09-01 – 2023-03-01|
Our group has a number of ongoing projects related to enzyme evolution of cold adaptation (coordinated by Prof. Åqvist).  Here different psychrophilic enzymes are considered, many of which have no experimental structure available. In these cases we will model accurate structures with machine learning based methods, namely Alphafold 2. We then move on to simulate cold-adapted enzymes and their warm-adapted homologues with extensive molecular dynamics (MD) simulations, mainly performed with the GROMACS package. The experimental design of these simulations includes running several mutants, under different conditions (temperature), and a subsequent integrated analysis including clustering and classification algorithms. We could ultimately generate and train a ML model to relate structural differences between enzymes to the observed temperature dependance of catalysis. From all the above, the Berzelius AI/ML computing cluster could be a great advantage for these projects.
Another active area in the lab (coordinated by Assoc Prof. Gutiérrez-de-Terán) is related to the mechanistic characterization of signal transduction through membrane proteins, with emphasis on the G-protein-coupled receptors (GPCRs) systems. We have currently ongoing projects in collaboration with experimental groups of Leiden and Florida universities, to characterize the mechanistic effect of cancer somatic mutations identified by our collaborations in both the receptor and the intracellular signaling G- protein found in cancer. These mutations are characterized using atomistic MD simulations of different systems (i.e., GPCR, G-protein, and the complex of both, all under different conditions (presence/absence of allosteric and orthosteric ligands, different mutant combinations, etc) . For these simulations, we make use of our PyMemDyn protocol to simulate membrane proteins with the GROMACS package. Analogous to what was described in the previous project, a subsequent integrated analysis could ultimately generate and train a ML model to identify pathogenic mutations that modify the physiological profile of signal transduction.
A joint methodological activity of Åqvist and Gutiérrez-de-Terán is the generation of large scale virtual screening based on free energy perturbation simulations.  Briefly, this approach allows accurate estimation of relative binding free energy (and thus affinities) of pairs of compounds within a dataset of drug candidates. The initial design of the map of pairwise comparisons is done on the basis of algorithms related to chemical similarity of the compound dataset, but in this application we aim to train ML learning algorithms with the FEP results of these initial selections (on a classical benchmark of targets). The code for performing such calculations (https://github.com/qusers/Q6) has been rewritten to compile on CUDA compilers and run on GPUs, which we want to test and train along with this application.
 Åqvist, J; Isaksen, G.V.; Brandsdal, BØ. "Computation of enzyme cold adaptation." Nature Reviews Chemistry 1.7 (2017): 1-14.
 Wei, S.; Thakur, N.; Ray, A.P.; Jin, B.; Obeng, S.; McCurdy, C.R.; McMahon, L.R.; Gutiérrez-de-Terán, H.; Eddy, M.T.; Lamichhane, R. "Slow Conformational Dynamics of the Human A2A Adenosine Receptor are Temporally Ordered." Structure 30 (2022) 329
 Jespers, , et al. "QligFEP: an automated workflow for small molecule free energy calculations in Q." Journal of cheminformatics 11.1 (2019): 1-16.