Network Generators in konnektor#

In this notebook is an introduction to the network topologies available in konnektor.

Generating randomized input data#

First we generate a randomized toy input dataset.

Note that these only work in combination with each other! A real world example can be found in the [openfe example notebooks]((OpenFreeEnergy/ExampleNotebooks).

[2]:

from konnektor.visualization import draw_ligand_network

from konnektor.utils.toy_data import build_random_dataset

components, mapper, scorer = build_random_dataset(n_compounds=30, rand_seed=None)

Feel free to play with the parameters! n_compounds controls the number of compounds, and rand_seed can be used to change the random seed.

Star Network#

The Star Network is one of the most edge efficient layouts, connecting all the Components to one central Component.

The default algorithm is as follows:

  1. Constructs all the possible connections between the Components.

  2. Select the Component is the best average connection score as the central Component.

  3. Only retain connections to this central Component.

While this network is most efficient in terms of edges, it does not necessarily have the most efficient set of connections. This is because a central Component has to be chosen, which is often a compromise for all Components (i.e. it may not be the easier Component to transform to, it just happens to have the best average connection score).

From a robustness point of view, the Star Network will immediately be disconnected if one connection fails. However the loss of Components is very limited, as only one Component is lost per failure.

[3]:

from konnektor.network_planners import StarNetworkGenerator

ligand_network_planner = StarNetworkGenerator(mappers=mapper, scorer=scorer, n_processes=1)

[4]:

star_network = ligand_network_planner.generate_ligand_network(components)
star_network.name = "Star Network"
star_network.name

[4]:

'Star Network'

[5]:

draw_ligand_network(network=star_network, title=star_network.name);
../_images/tutorials_network_generation_algorithms_7_0.png

Twin Star Network#

The Twin Star Network is an expansion to the Star Network. It can be described as multiple star networks that are combined.

The default algorithm is as follows:

  1. Constructs all the possible connections between the Components.

  2. The n_centers Components with the best average connection scores are selected an placed into the center of the network.

  3. All the components are connected to the selected central Components, resulting in \(n_{connections} = n_{centers}*(n_{Components}-n_{centers})\).

Note: The n_centers option allows you to change the Twin Star to a Triplet Star Network or more.

This approach leads to double the number of connections compared to the Star Network and therefore has a higher cost to complete. However, it builds a lot of graph cycles which add redundancy and allow for cycle closure analysis.

Similarly to the Star Network, node connectivity is centralized around the n_centers. This means, that the selection of the central ligands is very important.

[26]:

from konnektor.network_planners import TwinStarNetworkGenerator

ligand_network_planner = TwinStarNetworkGenerator(
    mappers=mapper, scorer=scorer, n_centers=2
)

[27]:

twin_star_network = ligand_network_planner.generate_ligand_network(components)
twin_star_network.name = "Twin Star Network"
twin_star_network.name

[27]:

'Twin Star Network'

[28]:

draw_ligand_network(network=twin_star_network, title=twin_star_network.name);
../_images/tutorials_network_generation_algorithms_11_0.png

Starry Sky Network#

The StarrySkyNetworkGenerator is an advanced network algorithm that clusters the provided Components and builds up a network from these clusters.

The network algorithm follows the following steps:

  1. Component clustering:

    1. Translate the Molecules into Morgan Fingerprints. (default)

    2. Cluster the Morgan Fingerprints with HDBSCAN. (default)

  2. Build Sub-Star Networks for each Cluster using the StarNetworkGenerator.

  3. Concatenate the Sub-Star Networks to the final Starry Sky Network, with 3 connections per cluster pair using the MSTConcatenator.

This approach allows networks to be built with portions of the Components connected to a central Component that is most representative of a given feature. This can lead to an improved graph score.

This does come at the cost of a higher number of connections (compared to a Star Network), but not as much as the number which would be generated by a Twin Star Network with the same number of Components.

[29]:

from sklearn.cluster import HDBSCAN
from scikit_mol.fingerprints import MorganFingerprintTransformer
from konnektor.network_tools import ComponentsDiversityClusterer
from konnektor.network_planners import StarrySkyNetworkGenerator

ligand_network_planner = StarrySkyNetworkGenerator(
    mappers=mapper,
    scorer=scorer,
    clusterer=ComponentsDiversityClusterer(
        featurize=MorganFingerprintTransformer(),
        cluster=HDBSCAN(metric="jaccard", min_cluster_size=3, alpha=1 / 2048),
    ),
    n_processes=1,
)

[30]:

starry_sky_network = ligand_network_planner.generate_ligand_network(components)
starry_sky_network.name = "Starry Sky Network"
starry_sky_network.name

/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/sklearn/cluster/_hdbscan/hdbscan.py:722: FutureWarning: The default value of `copy` will change from False to True in 1.10. Explicitly set a value for `copy` to silence this warning.
  warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(

[30]:

'Starry Sky Network'

[31]:

draw_ligand_network(network=starry_sky_network, title=starry_sky_network.name);
../_images/tutorials_network_generation_algorithms_15_0.png

Minimal Spanning Tree#

The MinimalSpanningTreeNetworkGenerator, builds a minimal spanning tree (MST) network for a given set of Components.

For the MST algorithm the Kruskal Algorithm is used.

The MST algorithm gives the optimal graph score possible and the minimal required set of connections to make sure all the Component nodes are connected.

This makes the MST Network very efficient. However, the MST is not very robust as one failed connection wwill lead to the network being immediatly disconnected.

[12]:

from konnektor.network_planners import MinimalSpanningTreeNetworkGenerator

ligand_network_planner = MinimalSpanningTreeNetworkGenerator(
    mappers=mapper, scorer=scorer
)

[32]:

mst_network = ligand_network_planner.generate_ligand_network(components)
mst_network.name = "Min Spanning Tree Network"
mst_network.name

/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/sklearn/cluster/_hdbscan/hdbscan.py:722: FutureWarning: The default value of `copy` will change from False to True in 1.10. Explicitly set a value for `copy` to silence this warning.
  warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(

[32]:

'Min Spanning Tree Network'

[33]:

draw_ligand_network(network=mst_network, title=mst_network.name);
../_images/tutorials_network_generation_algorithms_19_0.png

N-Node Edges Network#

The N-Node Edges Network tries to add more redundancy to the MST Network to improve the robustness.

The network algorithm follows the following steps:

  1. Build an MST Network.

  2. Add the best scoring connections trying to reach target_node_connectivity for each Component.

[41]:

from konnektor.network_planners import NNodeEdgesNetworkGenerator

ligand_network_planner = NNodeEdgesNetworkGenerator(
    mappers=mapper, scorer=scorer, target_component_connectivity=3
)

[36]:

nedges_network = ligand_network_planner(components)
nedges_network.name = "N-Edges Network"
nedges_network.name

[36]:

'N-Edges Network'

[37]:

draw_ligand_network(network=nedges_network, title=nedges_network.name);
../_images/tutorials_network_generation_algorithms_23_0.png

Redundant MST#

The RedundantMinimalSpanningTreeNetworkGenerator is an approach that attempts to minimize the impact of connection failures in an MST-like network.

The network algorithm follows the following steps:

  1. Create n_redundancy times MST networks, removing the nodes already selected at each iteration.

  2. Build an overlay of all the MST networks genreated.

This increases the number of redundant edges, reducing the impact of failures.

[38]:

from konnektor.network_planners import RedundantMinimalSpanningTreeNetworkGenerator

ligand_network_planner = RedundantMinimalSpanningTreeNetworkGenerator(
    mappers=mapper, scorer=scorer, n_redundancy=2
)

[39]:

rmst_network = ligand_network_planner.generate_ligand_network(components)
rmst_network.name = "Redundant Min Spanning\n Tree Network"
rmst_network

[39]:

<LigandNetwork-2ffa4556e307c74bd781d00d861c7c00>

[40]:

draw_ligand_network(network=rmst_network, title=rmst_network.name);
../_images/tutorials_network_generation_algorithms_27_0.png

Cyclic Network#

The CyclicNetworkGenerator creates a network with internal cycles within the network. This can be useful in ensuring redundancy and allow for cycle closure analysis.

The network algorithm follows the following steps:

  1. For each node, the algorithm generates all cycles of size cycle_sizes and assigns a score to each cycle as the sum of all sub-scores.

  2. Next it selects the node_present_in_cycles best score perfoming and node diversity increasing (see below) cycles per node.

  3. From these a graph is constructed.

The node diversity is used to spread the cycle in the network equallly between all the Components.

The number of cylces, around each Component can be defined by component_present_in_cycles and allowed cylce size can be tweaked with cycle_sizes. For cycle_sizes you can either use an integer for providing an expected cycle size (e.g. 3) or a range of allowed cycle sizes (e.g. [3,4]).

This network topology has a well distributed connectivity between all Components, which increases the robustness but still allows for a better graph score then the Twin Star Network, as the connectivity distribution is biased not enforced.

[21]:

from konnektor.network_planners import CyclicNetworkGenerator

ligand_network_planner = CyclicNetworkGenerator(
    mappers=mapper, scorer=scorer, cycle_sizes=[3, 4], node_present_in_cycles=2
)

[22]:

cyclic_network = ligand_network_planner.generate_ligand_network(components)
cyclic_network.name = "Cyclic Network"
cyclic_network

[22]:

<LigandNetwork-d33b929a76937c43ff10c4fee72833ce>

[23]:

draw_ligand_network(network=cyclic_network, title=cyclic_network.name);
../_images/tutorials_network_generation_algorithms_31_0.png

Maximally connected network#

The MaximalNetworkGenerator builds a graph of all possible connections between a set of Components.

This approach is not very suitable for production use as it has the highest possible graph cost. However, it is very important as a precursor to creating all the other network types.

Note: the MaximalNetworkGenerator is parallelized and the number of CPUs can be given with n_processes. All other approaches in Konnektor benefit from this parallelization and you can use this parallelization with n_processes key word during class construction.

[44]:

import logging
from konnektor.network_planners import MaximalNetworkGenerator

ligand_network_planner = MaximalNetworkGenerator(
    mappers=mapper, scorer=scorer, n_processes=3
)

[46]:

max_network = ligand_network_planner.generate_ligand_network(components)
max_network.name = "Maximal Network"
max_network.name

/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(

[46]:

'Maximal Network'

[47]:

draw_ligand_network(network=max_network, title=max_network.name);
../_images/tutorials_network_generation_algorithms_35_0.png

Heuristic Maximal Network#

The HeuristicMaximalNetworkGenerator builds a network with n_samples connections per Component for a given set of Components.

Similar to MaximalNetworkGenerator, this network can be used as a precursor to other network types.

Note: the HeuristicMaximalNetworkGenerator is parallelized and the number of CPUs can be given with n_processes.

[48]:

import logging
from konnektor.network_planners import HeuristicMaximalNetworkGenerator

ligand_network_planner = HeuristicMaximalNetworkGenerator(
    mappers=mapper,
    scorer=scorer,
    n_samples=10,
    n_processes=3,
)

[49]:

hmax_network = ligand_network_planner.generate_ligand_network(components)
hmax_network.name = "Heuristic Maximal\nNetwork"
hmax_network.name

/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(

[49]:

'Heuristic Maximal\nNetwork'

[50]:

draw_ligand_network(network=hmax_network, title=hmax_network.name);
../_images/tutorials_network_generation_algorithms_39_0.png

Custom Diversity Cluster Network#

This class is the powers the Starry Sky Network and can be used to create custom clustered network types. Here we demonstrate how it can be used to use Cyclic Networks instead of Star Networks.

The ClusteredNetworkGenerator implements the following algorithm:

  1. Cluster Components with the clusterer obj.

  2. Build sub-networks in the clusters using the sub_network_planners.

  3. Concatenate all sub-networks using the concatenator in order to build the final network.

[51]:

from konnektor.network_planners import ClusteredNetworkGenerator
from konnektor.network_planners import CyclicNetworkGenerator
from konnektor.network_planners import MstConcatenator

sub_network_planner = CyclicNetworkGenerator(mappers=mapper, scorer=scorer)
network_concatenator = MstConcatenator(mappers=mapper, scorer=scorer)
ligand_network_planner = ClusteredNetworkGenerator(
    sub_network_planners=sub_network_planner, concatenator=network_concatenator
)

[54]:

div_network = ligand_network_planner.generate_ligand_network(components)
div_network.name = "Diversity Clustering Network"
div_network

/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(
/Users/atravitz/micromamba/envs/konnektor/lib/python3.12/site-packages/gufe/components/explicitmoleculecomponent.py:74: UserWarning: RDKit does not preserve Mol properties when pickled by default, which may drop e.g. atom charges; consider setting `Chem.SetDefaultPickleProperties(Chem.PropertyPickleOptions.AllProps)`
  warnings.warn(

[54]:

<LigandNetwork-5d4cef2580b0e942065c0f7be1fbcde8>

[55]:

draw_ligand_network(network=div_network, title=div_network.name);
../_images/tutorials_network_generation_algorithms_43_0.png

Summary#

Let’s generate a nice summary figure, with all the shown network layouts.

[56]:

networks = [
    star_network,
    twin_star_network,
    starry_sky_network,
    nedges_network,
    rmst_network,
    cyclic_network,
    div_network,
    hmax_network,
    max_network,
]

networks = list(sorted(networks, key=lambda x: len(x.edges)))

import numpy as np
from matplotlib import pyplot as plt

fig, axes = plt.subplots(ncols=5, nrows=2, figsize=[9 * 4, 2 * 9])
axes = np.array(axes).flat
fs = 22
for ax, net in zip(axes, networks):
    draw_ligand_network(network=net, title=net.name, ax=ax, node_size=1500, fontsize=40)
    ax.axis("off")
axes[-1].axis("off")
fig.tight_layout()
../_images/tutorials_network_generation_algorithms_45_0.png 
[58]:

# NBVAL_SKIP
# uncomment to save this image locally

# fig.savefig("./network_layouts.png", dpi=400)

[ ]: