MobleyLab/HiMap

Contents of this file

HiMap Introduction

HiMap includes design generation based on statistical optimality. Alchemical free energy calculations hold increasing promise as an aid to drug discovery efforts. However, applications of these techniques in discovery projects have been relatively rare, partly because of the difficulty of planning and setting up calculations. The lead optimization mapper (LOMAP) was introduced as an automated algorithm to plan relative free energy calculations between potential ligands. LOMAP was further developed to be based on free, open-source APIs such as RDKit. HiMap now includes clustering of ligand series, and optimization of free energy perturbation networks.

Authors

Contact for HiMap:

Mary Pitman mpitman@uci.edu
David Mobley dmobley@uci.edu

Predecessor authors of LOMAP:

Gaetano Calabro' gcalabro@uci.edu
Mark Mackey
Lester Hedges
Antonia S J S Mey
Jenke Scheen
David Mobley

Installation

To install HiMap with LOMAP included, build the conda environment and install from file:

https://github.com/MobleyLab/HiMap/tree/main/devtools/conda-envs/himap_env.yml

with:

conda env create -f himap_env.yml

Usage

Example scripts are included for various purposes:

To run HiMap with optimization
python examples/example_optimize.py
To read in scores and optimize
python examples/example_optimize_read_data.py
For a basic LOMAP run without optimization
python examples/example.py
For generating radial graphs with a hub
python examples/example_radial.py

To run LOMAP without optimization, as a commandline tool:

lomap test/basic/

If you would rather use the API directly:

To generate optimal designs, try:

cd examples/

import lomap
import himap

#-------------------------------------------------------#
# Generate similarity scores.
#-------------------------------------------------------#
# Read molecules from test directory.
db_mol = lomap.DBMolecules('../test/radial/', output=True, radial=True)
    
# Generate the strict and loose symmetric similarity score matrices.
strict, loose = db_mol.build_matrices()
    
# Convert the similarity matrix to numpy array
sim_np = strict.to_numpy_2D_array()

# Clean data if Lomap produces rare error. If score is NaN, replace with 0.0
n_arr = himap.clean_NaN(sim_np)

#-------------------------------------------------------#
# Clustering.
#-------------------------------------------------------#
# Create ID_list from db_mol prior to clustering.
ID_list = himap.db_mol_IDs(db_mol, n_arr)

# Perform clustering.
#   sub_arr, sub_ID:   the n_arr and ID_list subdivided by clusters
#   selected_clusters: user selected clusters during interaction.
sub_arr, sub_ID, selected_clusters = himap.cluster_interactive(n_arr, ID_list)

#-------------------------------------------------------#
# Optimization.
#-------------------------------------------------------#
# Example reference ligand.
ref_ligs = ['ejm_31']

# Send the user selected clusters for optimization.
himap.clusters2optimize(sub_arr, sub_ID, clusters2optim = selected_clusters, ref_ligs=ref_ligs)

To generate optimal designs using external scores or weights, try:

import himap

#-------------------------------------------------------#
# Define input files, read data.
#-------------------------------------------------------#
# Input files for weight scores and ligand names.
sim_scores_in = '../test/optimize/sim_scores.csv'
IDs_in = '../test/optimize/mol_names.txt'

# Read files, clean any potential NaN scores.
#   Added optional parameter:
#             delimiter: default is ','
n_arr, ID_list = himap.read_data(sim_scores_in, IDs = IDs_in)

#-------------------------------------------------------#
# Clustering.
#-------------------------------------------------------#
# Perform clustering.
#   sub_arr, sub_ID:   the n_arr and ID_list subdivided by clusters
#   selected_clusters: user selected clusters during interaction.
sub_arr, sub_ID, selected_clusters = himap.cluster_interactive(n_arr, ID_list)

#-------------------------------------------------------#
# Optimization.
#-------------------------------------------------------#
# Example reference ligands.
ref_ligs = ['mol_0', 'mol_1', 'mol_2', 'mol_3', 'mol_4']

# Send the user selected clusters for optimization.
himap.clusters2optimize(sub_arr, sub_ID, clusters2optim = selected_clusters,
                        ref_ligs=ref_ligs, num_edges = '2n', optim_types = ['A', 'D']
                        )

To generate original LOMAP designs, try:

import lomap

# Generate the molecule database starting from a directory containing .mol2 files

db_mol = lomap.DBMolecules("python string pointing to a directory with mol2 files", output=True)

    #More graphing options:
    # Use the complete radial graph option. The ligand with the most structural similarity to all of the others will be picked as the 'lead compounds' and used as the central compound.
    db_mol = lomap.DBMolecules("python string pointing to a directory with mol2 files", output=True, radial=True)

    # Use a radial graph with a manually specified hub compound
    db_mol = lomap.DBMolecules("python string pointing to a directory with mol2 files", output=True, radial=True, hub=filename.mol2)

    # Use a radial graph with a manually specified hub compound and fast graphing option
    #the fast graphing option create the initial graph by connecting the hub ligand with the possible surrounding ligands and add surrounding edges based on the similarities accoss surrounding nodes
    db_mol = lomap.DBMolecules("python string pointing to a directory with mol2 files", output=True, radial=True, hub=filename.mol2, fast=True)

# Calculate the similarity matrix betweeen the database molecules. Two molecules are generated
# related to the scrict rule and loose rule 

strict, loose = db_mol.build_matrices()

# Generate the NetworkX graph and output the results
nx_graph = db_mol.build_graph() 


# Calculate the Maximum Common Subgraph (MCS) between 
# the first two molecules in the molecule database 
# ignoring hydrogens and depicting the mapping in a file
    
MC = lomap.MCS.getMapping(db_mol[0].getMolecule(), db_mol[1].getMolecule(), hydrogens=False, fname='mcs.png')


# Alchemical transformation are usually performed between molecules with
# the same charges. However, it is possible to allow this transformation
# manually setting the electrostatic score for the whole set of molecules 
# producing a connected graph. The electrostatic scrore must be in the 
# range [0,1]


db_mol = lomap.DBMolecules("python string pointing to a directory with mol2 files", output=True, ecrscore=0.1)
strict, loose = db_mol.build_matrices()
nx_graph = db_mol.build_graph()

Requirements

RDKit Release > 2021
NetworkX
Matplotlib
python >= 3.8
R
rpy2=3.4.5
kneed=0.7.0
scikit-learn=0.23.2
scipy

Troubleshooting

Why is optimization not finding a random seed design?

Check that similarity scores are not zero, nearly zero, or in large part near zero:
- If the similarity scores are non-zero but very dissimilar, try decreasing the neighbor distance cutoff value (epsilon) selected during clustering.
- If the similarity scores are zero or near zero, verify if relative binding free energy calculations or relative alchemical transformations should be run for this set.
- Alternatively, if the ligands are chemically similar but break the rules for similarity in LOMAP, consider using another similarity metric. For example, ring breaking results in a score of zero with the LOMAP version included here. If your transformation method performs ring breaking well, use another similarity score or refine the scoring in LOMAP.
For designs with about 40 ligands and more, an edge count near the number of ligands may be too sparse for a weighted random invertible matrix to be found for your set of ligands. Increase the edge count. The coming version of HiMap will improve upon this feature.

MobleyLab / HiMap