---
type: slide
slideOptions:
theme: white
---

# GROMACS in Google.Colab

---

## Making it rain


OpenMM simulations are in Google.Colab. Why not GROMACS?

---

## Workflow

- Install everything (python libraries, etc.)
- Set user-defined options (GROMACS version, force fields, mutants to simulate, way to fix structure, how to setup protonation states)
- Load GROMACS from preinstalled package
- Run everything
- Analysis

---

## Current choices: Input

Input structure in one of the forms:
- PDB ID
- Upload from Google Drive
- Upload from computer

TODO:
- Add `AlphaFold`

---

## GROMACS

Currently available:
- 2021.5
- 2022.3

TODO:
- Test performance for different installation, run flags

---

## Force fields
- CHARMM36
- AMBER14SB
- AMBER99ILDN

---

## Structure cleaning

- Currently done with `pdbfixer`
- Easy to add `Modeller`

---

## Simulation of mutants
- The user can provide a list of mutants to simulate. Structures of the mutants are generated using `Modeller`

---

## Setting up protonation states
- Currently done with [`pKa-ANI`](https://github.com/isayevlab/pKa-ANI)

- Easy to add [`propka`](https://propka.readthedocs.io/en/latest/index.html)

---

## Run simulations and analysis

- Currently only RMSD, but easy to add RMSF, clustering and other types

---

## Problems
- GROMACS efficiency (60 ns/day compared 150 ns/day for same size with `OpenMM`)
- Hard-drive memory

---

## Ideas for the future

- Add membranes
- Add ligands
- Add alanine scan
- Add FE simulations for ligands

---

## What to show in the manuscript?
- Benchmark
- What else?

---

# GROMACS + AlphaFold2

digraph chargesetter {
    pad= 0.1
    splines= line
    nodesep= 1.2
    ranksep= 1
    fontname= "Sans-Serif"
    fontsize= 13
    fontcolor= "#2D3436"

    node [shape = box
        style = rounded
        fixedsize = true
        width = 1.4
        height = .7
        fontname = "Sans-Serif"
        fontsize = 13
        fontcolor = "#2D4436"]
    
    sequence [color="red", label="sequence"]
    structure [color="red", label="structure"]
    trajectory [color="red", label="trajectory"]
    analysis [color="red", label="analysis"]
    cg [color="k", label="coarse grained"]
    parameters [label="parameters:\nions, box,\nsolvent, mdp"]
    FF [color="red"]
    Amber99ildn [color="red"]
    
    sequence->structure [label="AF2/Colab??"]
    structure->protonation [label="propka"]
    protonation->pdb2gmx [label="send optimal\nprotonation"]
    structure->FF
    structure->parameters
    structure->membrane [label="how to know if\nthe protein is\nmembrane"]
    ligands
    cg
    FF->Amber99ildn
    FF->Charmm36 [label="predefined MDPs"]
    pdb2gmx->trajectory
    Amber99ildn->trajectory
    Charmm36->trajectory
    membrane->trajectory
    trajectory->analysis
    analysis->rmsd
    analysis->rmsf
}

- Install GROMACS
- We agreed to set up GROMACS from docker first and adjust the GMX part of the workflow
- Install AF2
- The problem here is with large databases needed to run inference
- Maybe it is easier to send queries to [trRosetta](https://yanglab.nankai.edu.cn/trRosetta/)
- Basic workflow (seq->str->def prot, par->Amber->traj->analysis): [link](https://colab.research.google.com/drive/1hFuZPZaD-zCUo4HFy3257q9VSzIxgkQy?usp=sharing)
- Add protonation
- Add CHARMM36
- Add parameters
- Add membrane
- Add CG/ligands