SOMMD

Self Organising Map for analysis of Molecular Dynamics data

Features

SOMMD provides functions to processes data from Molecular Dynamics simulations using Self Organising Maps. Features include the ability to read different input formats such as pdb, gro, dcd and xtc files. Trajectory can be analysed to identify groups of important frames.

Installing SOMMD

For the majority of users we recommend the use of the last stable release available from CRAN. To install from within R issue the command:

#!r
install.packages("SOMMD", dependencies=TRUE)

Tutorials

The package includes three tutorial notebooks located in the inst/Rmd folder, demonstrating different applications of the SOMMD package:

1. Clustering of MD trajectories
   This tutorial introduces the use of Self-Organizing Maps (SOM) for analyzing the conformational landscape of a protein from molecular dynamics (MD) simulations. It demonstrates how SOMMD can group similar structures, visualize macrostates, and extract representative conformations.

2. Analysis of Pathways in Protein Unfolding Simulations
   This tutorial applies SOM to Steered MD (SMD) simulations to investigate protein unfolding pathways. It showcases how SOM can classify different unfolding states and track pathways sampled across simulations, enabling a comparative analysis of unfolding mechanisms.

3. Transition Network Analysis in Ligand-Protein Metadynamics Simulations
   This tutorial focuses on ligand binding dynamics using metadynamics simulations. It explains how to train a SOM on ligand-protein interaction data and build a transition network to analyze binding pathways, providing insights into the complexity of the binding process.

List of Functions and Their Usage

The table below provides a complete list of functions available in the SOMMD package, categorized by their role in the analysis workflow. The functions are grouped into Data Import and Preprocessing, which are essential for all analyses, and those specific to Conformational State Analysis, Pathway Analysis, and Network Analysis. This structured overview helps users understand how different functions contribute to each stage of the analysis.

Data Import

Function Name	Description
read.struct	Reads molecular structure files (PDB/GRO) and returns a struct object encapsulating atomic coordinates, box dimensions, and topological information.
read.trj	Reads trajectory files (XTC/DCD) and returns a trj object containing the 3D coordinate array, topology, and frame indices.
cat.trj	Concatenates multiple trajectory files into a single trj object, ensuring consistency across simulations.
trj2xyz	Converts the 3D trajectory array into a 2D XYZ coordinate matrix for further processing or visualization.
print.struct	Prints a summary of a struct object, highlighting key features of the imported structure.
print.trj	Prints a summary of a trj object, providing basic information about the trajectory data.

Preprocessing

Function Name	Description
fit.trj	Aligns a trajectory to a reference structure using the Kabsch algorithm (via bio3d), facilitating structural comparisons.
calc.distances	Computes distance matrices from trajectory data, serving as input for SOM training.
native.cont	Selects native contact distances from a reference structure or a trajectory frame based on a user-specified distance cutoff.

Analysis of Conformational States

Function Name	Description
cluster.representatives	Identifies representative frames for each cluster, providing key conformations that summarize the cluster’s behavior.
silhouette.profile	Calculates the silhouette profile for a given number of clusters, helping evaluate the quality of the clustering.
silhouette.score	Computes silhouette scores across a range of cluster numbers to assist in determining the optimal cluster partition.
neur.population	Computes the per-neuron population (number of frames assigned to each neuron), which can be used to assess cluster significance.
neur.representatives	Determines the representative frame for each neuron by selecting the frame closest to the neuron’s codebook vector.
average.neur.property	Computes the average value of a given property for each neuron within the SOM, aiding in the characterization of neuron-specific features. Useful to visualize properties.
remap.data	Maps new data onto a pre-trained SOM, allowing the extension of the analysis to additional simulations or datasets.
som.add.circles	Adds circles to a SOM plot—sized proportionally to a selected property—to enhance the graphical representation of the SOM.

Pathway Analysis

Function Name	Description
trace.path	Traces the path of a given trajectory through the SOM, allowing the reconstruction of the sequence of visited neurons.
cluster.pathways	Clusters conformational pathways based on time-dependent or independent schemes, facilitating the exploration of dynamic transitions.

Network Analysis

Function Name	Description
comp.trans.mat	Computes the transition matrix between SOM neurons, quantifying the probability of transitions between neurons in the simulation.
matrix2graph	Converts a transition matrix into an igraph object, enabling network visualization of state transitions between SOM neurons.
map.color	Maps a numeric property vector to a color scale based on a provided palette, useful for annotating visual outputs.

Citation

To cite SOMMD in publications use:

Motta, S., Callea, L., Bonati, L., Pandini, A. (2022). “PathDetect-SOM: A Neural Network Approach for the Identification of Pathways in Ligand Binding Simulations.” Journal of Chemical Theory and Computation, 18(3), 1957-1968. doi:10.1021/acs.jctc.1c01163 https://doi.org/10.1021/acs.jctc.1c01163.