In silico structural studies
of biological molecules
Supercomputing
challenges &
requirements
Emiliano De Santis, Giovanni La Penna,
Velia Minicozzi, Silvia Morante,
Giancarlo Rossi, Francesco Stellato
Dipartimento di Fisica & I.N.F.N.
Roma Tor Vergata
2018-2020
Large scale computing
@
2018-2020
Large Scale Computing @ INFN
Roma, February 13th , 2017
Outline
-- Structural Biology & HPC
- Protein aggregation
Neurodegenerative diseases
- Antibody-protein interactions
Tumour theranostic
- Ab initio observables calculations
X-ray spectra
Diffraction patterns
-- Computational Requirements
- Challenges & needs for 2018-2020
Classical MD
Ab initio calculations
2018-2020
Large Scale Computing @ INFN
Roma, February 13th , 2017
Classical Molecular Dynamics
Antibody-ligand
Fibril formation
5.000 (protein)
+
150.000 (water)
atoms
1000 (protein)
+
100.000 (water)
atoms
500 ns
(MD simulations)
> 100 ns
(MD simulations)
Software: GROMACS, LAMMPS
Sevigny et al. Nature 537 (2016)
Minicozzi et al. JBC 289 (2014); Carbonaro et al. BBA 1864 (2016)
De Santis et al. J Phys Chem B 119 (2105); Stellato et al. EBJ 43 (2014)
2018-2020
Large Scale Computing @ INFN
Roma, February 13th , 2017
Ab initio calculations
Metal ions–peptides interactions
MD simulations on fragments
of fibril-forming peptides (Aβ, prion)
100 (protein)
+
500 (water)
atoms
Giannozzi et al. Metallomics 4 (2012)
Software: QuantumESPRESSO
2018-2020
Large Scale Computing @ INFN
Roma, February 13th , 2017
Ab initio observable calculation
X-ray Absorption spectrum
Final state with
a core-hole
Initial state Incident photon
no core-hole
energy
X-ray Diffraction pattern
Charge density of crystalline molecules
calculated using Density Functional Theory
can be compared with experimental data
  2 2 Ze 2 
  i  E i
 
2m
r


 2 2

   V  f  E f
 2m

Density Functional Theory provides an
accurate description of the atomic potential
La Penna et al. J Chem Phys 143 (2015)
Stellato et al. IUCrJ 1 (2014)
2018-2020
Large Scale Computing @ INFN
Roma, February 13th , 2017
HPC Challenges for 2018-2020
Classical MD on systems with > 106 atoms
- Many A peptides  aggregation and interaction with small molecules
- Antibodies interaction with A peptides and tumour proteins
>107 core hours
QM simulations of systems with up to 1000-2000 atoms
- A few A1-40 peptides (400 atoms,  1200 electrons)
in complex with Cu(II) in a water environment
- XAS and XRD data calculation
 108 core hours
2018-2020
Large Scale Computing @ INFN
Roma, February 13th , 2017
HPC Requirements
QuantumESPRESSO tests
DFT PW calculation on a 77 atoms system
Galileo 8x16 cores = 128 cores = 7.4 s/step
Marconi A2 2x68 cores = 136 cores = 15.8 s/step
The key factor is the CPU clock
Marconi A2 @ 1.4 GHz, Galileo @ 2.4 GHz
Marconi A2 runs 2 times slower than Galileo
Machine desiderata
Core hours: 107
CPU: > 2.4 GHz
RAM: > 100 GB/node
Hard-Disk Space: > 100 TB
2018-2020
Large Scale Computing @ INFN
Roma, February 13th , 2017
People
• Prof. Silvia Morante
• Prof. Giancarlo Rossi
• Dr. Velia Minicozzi
• Dr. Francesco Stellato
• Dr. Emiliano De Santis
• Dr. Giovanni La Penna
The Biophysics Group in Tor Vergata
http://biophys.roma2.infn.it/
Large Scale Computing @ INFN
Roma, February 13th , 2017
Thank you for the attention!
???
2018-2020
Questions ???