🪄 Predict Protein Structure with Precision 🪄

Table of Contents

• Overview
• Installation
• Tutorial
  • 1. Input Preparations
  • 2. Cluster Assignment
  • 3. Cluster Trajectories
  • 4. Cluster Normalization
  • 5. Similarity Calculations
  • 6. Representative Frames
• Further Reading

Overview

Protein structures prediction is important because the accuracy of protein structures influence how our understanding of its function and its interactions with other molecules, which can help to design new drugs to target specific molecular interactions. Protein Retrieval via Integrative Molecular Ensembles (PRIME) is a novel algorithm that predicts the native structure of a protein from simulation or clustering data. This repo contains six different ways of determining the native structure of biomolecules from simulation or clustering data. These methods perfectly mapped all the structural motifs in the studied systems and required unprecedented linear scaling.

Installation

PRIME requires Python 3.6+ and the following packages: MDAnalysis, numpy, and matplotlib. bash git clone https://github.com/mqcomplab/PRIME.git cd PRIME

Tutorial

The following tutorial will guide you through the process of determining the native structure of a biomolecule using the PRIME algorithm. If you already have clustered data, you can skip to Step 4.

1. Input Preparations

Preparation for Molecular Dynamics Trajectory

Prepare a valid topology file (e.g. .pdb, .prmtop), trajectory file (e.g. .dcd, .nc), and the atom selection. This step will convert a Molecular Dynamics trajectory to a numpy ndarray. Make sure the trajectory is already aligned and/or centered if needed!

Step-by-step tutorial can be found in the scripts/inputs/preprocessing.ipynb.

2. Cluster Assignment

In this example, we will use k-means clustering to assign labels to the clusters and the number of clusters will be 20. Any clustering method can be used as long as the data is clustered (e.g. DBSCAN, Hierarchical Clustering). Please check out MDANCE for more clustering methods!

scripts/nani/assign_labels.py will assign labels to the clusters using k-means clustering

# System info - EDIT THESE
input_traj_numpy = '../../example/aligned_tau.npy'
N_atoms = 50
sieve = 1

# K-means params - EDIT THESE
n_clusters = 20
output_dir = 'outputs'

Inputs

System info

input_traj_numpy is the numpy array prepared from step 1.
N_atoms is the number of atoms used in the clustering, should be same as atom selection in step 1.
sieve takes every sieveth frame from the trajectory for analysis.

k-means params

n_clusters is the number of clusters for labeling.
output_dir is the directory where the output files will be saved.

Execution

bash python assign_labels.py

Outputs

1. csv file containing the cluster labels for each frame.
2. csv file containing the population of each cluster.

3. Cluster Trajectories

scripts/outputs/postprocessing.ipynb will use the indices from last step to extract the designated frames from the original trajectory for each cluster.

4. Cluster Normalization

With already clustered data, scripts/normalization/normalize.py will normalize the trajectory data between $[0,1]$ using the Min-Max Normalization.

# System info - EDIT THESE
input_top = '../../example/aligned_tau.pdb'
unnormed_cluster_dir = '../clusters/outputs/clusttraj_*'
output_base_name = 'normed_clusttraj'
atomSelection = 'resid 3 to 12 and name N CA C O H'
n_clusters = 10

Inputs

System info

input_top is the topology file used in the clustering.
unnormed_cluster_dir is the directory where the clustering files are located from step 3.
output_base_name is the base name for the output files.
atomSelection is the atom selection used in the clustering.
n_clusters is the number of clusters used in the PRIME. If number less than total number of cluster, it will take top n number of clusters.

bash python normalize.py

Outputs

1. normed_clusttraj.c*.npy files, normalized clustering files.
2. normed_data.npy, appended all normed files together.

5. Similarity Calculations

scripts/prime/exec_similarity.py generates a similarity dictionary from running PRIME.

• -h - for help with the argument options.
• -m - methods, pairwise, union, medoid, outlier (required).
• -n - number of clusters (required).
• -i - similarity index, RR or SM (required).
• -t - Fraction of outliers to trim in decimals (default is None).
• -w - Weighing clusters by frames it contains (default is True).
• -d - directory where the normed_clusttraj.c*.npy files are located (required)
• -s - location where summary file is located with population of each cluster (required)

Example

bash python ../../utils/similarity.py -m union -n 10 -i SM -t 0.1 -d ../normalization -s ../clusters/outputs/summary_20.txt

To generate a similarity dictionary using data in ../normalization (make sure you are in the prime directory) using the union method (2.2 in Fig 2) and Sokal Michener index. In addition, 10% of the outliers were trimmed. You can either python exec_similarity.py or run example above.

Outputs

w_union_SM_t10.txt file with the similarity dictionary. The result is a dictionary organized as followes: Keys are frame #. Values are [cluster 1 similarity, cluster #2 similarity, ..., average similarity of all clusters].

6. Representative Frames

scripts/prime/execrepframes.py will determine the native structure of the protein using the similarity dictionary generated in step 5.

• h - for help with the argument options.
• m - methods (for one method, None for all methods)
• s - folder to access for w_union_SM_t10.txt file
• i - similarity index (required)
• t - Fraction of outliers to trim in decimals (default is None).
• d - directory where the normed_clusttraj.c* files are located (required if method is None)

Example

bash python ../../utils/rep_frames.py -m union -s outputs -d ../normalization -t 0.1 -i SM

Outputs

w_rep_SM_t10_union.txt file with the representative frames index.

Further Reading

For more information on the PRIME algorithm, please refer to the PRIME paper. Please cite using CITATION.bib.

Fig 2. Six techniques of protein refinement. Blue is top cluster.

Funding

Research contained in this package was supported by the National Institute of General Medical Sciences of the National Institutes of Health under award number R35GM150620.