Invited talk at the SITOLA Seminar 2023, Telč, Czechia.
Created using Reveal.js.
Use the right arrow to start browsing the slides.
Exploring biomolecular dynamics through large-scale computer simulations
Michal H. Kolář
https://mhko.science/contact
1. biomolecules and their dynamics
2. MD simulations
3. supercomputers
4. case study: PDF
Biomolecules are biopolymers
biomolecules are not static
Activation pathway of a G protein-coupled receptor...
Martin Jensen – Solo Dance
feat.
Dylan Mayoral
Haeni Kim
brothers Jurečič
Input coordinates
Force field
V(i) = V(b) + V(nb)
Bonded interactions
V(b) = V(bonds) + V(angles) + V(torsions)
V(bond) = ½ k (d − d0)2
Non-bonded interactions
V(nb) = V(Coulomb) + V(vdw) ∝ r−1 + (r−12 − r−6)
Hands-on PDB and TOP file formats
Final data – a trajectory
number of atoms × number of trajectory frames
statistical ensemble of biomolecular conformations
typically MBs to TBs of data
foto: Mikael Kanerva link
foto: Fade Creative link
LUMI accelerated nodes
LUMI-G: 2560 nodes
1x AMD EPYC 7763 CPU
4x AMD MI250x GPU
512 GB RAM
LUMI-C: 1536 nodes
2x AMD EPYC 7763 CPU
256 − 1024 GB RAM
3,6 billion CZK
experimental devices
10 CZK
250 million CZK
120 million CZK
H Ψ = E Ψ
Schrödinger equation
amarican billionaire, philantropist, scientist
founder of D E Shaw Research
computer simulations of protein synthesis
McGrath, Černeková, Kolář: Binding of the peptide deformylase on the ribosome surface modulates the exit tunnel interior
PMID: 36335428
Strategy
Run MD simulations of two systems:
- with PDF
- without PDF
(challenging, time consuming)
Train a model that predicts presence/absence of PDF.
(routine, fast)
Find molecular parts that yield strong predictors.
(challenging, fast)
Prediction accuracy
