def test_plot_voronoi(self):
kmeans = KMeans(n_clusters=15)
kmeans.fit([data])
ax = plot_voronoi(kmeans, xlabel='x', ylabel='y')
assert isinstance(ax, SubplotBase)
def test_plot_voronoi():
kmeans = KMeans(n_clusters=15)
kmeans.fit([data])
ax = plot_voronoi(kmeans)
assert isinstance(ax, Subplot)
def cluster_kmeans(tica_dir, data_dir, traj_dir, n_clusters, lag_time):
clusterer_dir = "%s/clusterer_%dclusters.h5" % (tica_dir, n_clusters)
if (os.path.exists(clusterer_dir)):
print("Already clustered")
else:
print("Clustering by KMeans")
reduced_data = verboseload(data_dir)
trajs = np.concatenate(reduced_data)
clusterer = KMeans(n_clusters=n_clusters, n_jobs=-1)
clusterer.fit_transform(reduced_data)
verbosedump(clusterer, clusterer_dir)
def cluster_kmeans(tica_dir, data_dir, traj_dir, n_clusters, lag_time): clusterer_dir = "%s/clusterer_%dclusters.h5" %(tica_dir, n_clusters) if (os.path.exists(clusterer_dir)): print "Already clustered" else: print "Clustering by KMeans" reduced_data = verboseload(data_dir) trajs = np.concatenate(reduced_data) clusterer = KMeans(n_clusters = n_clusters, n_jobs=-1) clusterer.fit_transform(reduced_data) verbosedump(clusterer, clusterer_dir)
def cluster_msm(sequences,n_states, lag_times):
for n in n_states:
states = KMeans(n_clusters=n)
states.fit(sequences)
io.dump(states,str(n)+'n_cl.pkl')
ts=np.zeros(5)
for lag_time in lag_times:
msm = MarkovStateModel(lag_time=lag_time, verbose=False,n_timescales=5)
msm.fit(states.labels_)
ts1=msm.timescales_
ts=np.vstack((ts,ts1))
io.dump(msm,str(n)+'n_'+str(lag_time)+'lt_msm.pkl')
ts=np.delete(ts, (0), axis=0)
io.dump(ts,str(n)+'n_timescales.pkl')
def fit_protein_kmeans(yaml_file,mini=True,pca=False):
mdl_dir = yaml_file["mdl_dir"]
mdl_params = yaml_file["mdl_params"]
current_mdl_params={}
for i in mdl_params.keys():
if i.startswith("cluster__"):
current_mdl_params[i.split("cluster__")[1]] = mdl_params[i]
if mini:
current_mdl_params["batch_size"] = 100*current_mdl_params["n_clusters"]
kmeans_mdl = MiniBatchKMeans(**current_mdl_params)
else:
kmeans_mdl = KMeans(**current_mdl_params)
data = []
for protein in yaml_file["protein_list"]:
with enter_protein_mdl_dir(yaml_file, protein):
if pca:
tica_data = verboseload("pca_data.pkl")
else:
tica_data = verboseload("tica_data.pkl")
# get all traj
sorted_list = sorted(tica_data.keys(), key=keynat)
data.extend([tica_data[i] for i in sorted_list])
kmeans_mdl.fit(data)
kmeans_mdl_path = os.path.join(mdl_dir, "kmeans_mdl.pkl")
verbosedump(kmeans_mdl, kmeans_mdl_path)
return
def Kmeans_score(dataset, Max_clusters):
print(
"Start to analyze the dependence of inertia on the number of clusters\n"
)
scores_in = [] #the elbow indicates the good cluster number
scores_sc = [
] #s=b-a/max(a,b) a: The mean distance between a sample and all other points in the same class,
# b: The mean distance between a sample and all other points in the next nearest cluster.
scores_ch = [] # Variance Ratio Criterion, tightness of the cluster
scores_rt = [
] # Variance ratio, As the ratio inherently rises with cluster count,
# one looks for an “elbow” in the curve where adding another cluster does not add much new information, as done in a scree test
scores_db = [
] # Values closer to zero indicate a better partition. sum of cluster i and j diameter over the distance between cluster centroids i and j. smaller the better.
for i in range(Max_clusters - 2):
kmeans_model = KMeans(n_clusters=i + 2,
init='k-means++',
n_init=10,
max_iter=300,
tol=0.001,
precompute_distances='auto',
verbose=0,
random_state=None,
copy_x=True,
n_jobs=1).fit(dataset)
labels = kmeans_model.labels_
scores_in.append(kmeans_model.inertia_)
scores_sc.append(
metrics.silhouette_score(dataset[0], labels[0],
metric='euclidean'))
scores_ch.append(metrics.calinski_harabaz_score(dataset[0], labels[0]))
scores_rt.append(ssr_sst_ratio(dataset[0], labels[0]))
scores_db.append(metrics.davies_bouldin_score(dataset[0], labels[0]))
print("Done generating scores for " + str(Max_clusters) + " clusters\n")
return scores_in, scores_sc, scores_ch, scores_rt, scores_db
def cluster_project_wrapper(proj_folder,feature_dict,n_states):
if os.path.exists(proj_folder+"/assignments.pkl"):
return verboseload(proj_folder+"/cluster_mdl.pkl"),verboseload(proj_folder+"/assignments.pkl")
elif os.path.exists(proj_folder+"/cluster_mdl.pkl"):
cluster_mdl = verboseload(proj_folder+"/cluster_mdl.pkl")
else:
cluster_mdl = KMeans(n_clusters = n_states)
cluster_mdl.fit([feature_dict[i] for i in feature_dict.keys()])
assignments={}
for i in feature_dict.keys():
assignments[i] = cluster_mdl.transform([feature_dict[i]])
verbosedump(cluster_mdl,proj_folder+"/cluster_mdl.pkl")
verbosedump(assignments,proj_folder+"/assignments.pkl")
return cluster_mdl,assignments
def cluster_features(features, clusterer, n_clusters=8):
'''
Input
features : list of arrays, length n_trajs, each of shape (n_samples, n_features)
Output
clst : msmbuilder.cluster object, with attributes
cluster_centers_ : (n_clusters, n_features)
labels_ : list of arrays, each of shape (n_samples, )
'''
if clusterer == 'KMeans':
from msmbuilder.cluster import KMeans
clst = KMeans(n_clusters=n_clusters)
elif clusterer == 'KCenters':
from msmbuilder.cluster import KCenters
clst = KCenters(n_clusters=n_clusters)
elif clusterer == 'KMedoids':
from msmbuilder.cluster import KMedoids
clst = KMedoids(n_clusters=n_clusters)
elif clusterer == 'MiniBatchKMeans':
from msmbuilder.cluster import MiniBatchKMeans
clst = MiniBatchKMeans(n_clusters=n_clusters)
elif clusterer == 'MiniBatchKMedoids':
from msmbuilder.cluster import MiniBatchKMedoids
clst = MiniBatchKMedoids(n_clusters=n_clusters)
clusters = clst.fit_transform(features)
return clst
def cluster_project_wrapper(proj_folder, feature_dict, n_states):
if os.path.exists(proj_folder + "/assignments.pkl"):
return verboseload(proj_folder +
"/cluster_mdl.pkl"), verboseload(proj_folder +
"/assignments.pkl")
elif os.path.exists(proj_folder + "/cluster_mdl.pkl"):
cluster_mdl = verboseload(proj_folder + "/cluster_mdl.pkl")
else:
cluster_mdl = KMeans(n_clusters=n_states)
cluster_mdl.fit([feature_dict[i] for i in feature_dict.keys()])
assignments = {}
for i in feature_dict.keys():
assignments[i] = cluster_mdl.transform([feature_dict[i]])
verbosedump(cluster_mdl, proj_folder + "/cluster_mdl.pkl")
verbosedump(assignments, proj_folder + "/assignments.pkl")
return cluster_mdl, assignments
def cluster():
'''
This function perfomes K-means clustering on the tICA space and saves assignsment files for each trajectory.
Cluster centers are also saved at `microstate_centers.txt` file.
'''
cluster = KMeans(n_clusters=n_states,n_jobs=-1,verbose=0, max_iter=100, tol=0.0001,)
dataset, ev0, ev1 = [], [], []
print "Loading projected data..."
for i in tqdm(range(start_traj, end_traj+1)):
a = io.loadh('%s/traj%d_%s.h5' %(proj_path,i,traj_name))['arr_0']
a = a[:,0:2]
dataset.append(a)
ev0.extend(a[:,0])
ev1.extend(a[:,1])
print "Clustering %d datapoints..." %len(ev0)
cluster.fit(dataset)
for i in range(start_traj,end_traj+1):
np.savetxt('%s/assigns_%d.txt' %(out_path,i),np.array(cluster.labels_[i-start_traj]),fmt='%d')
np.savetxt('%s/microstate_centers.txt' %out_path,np.array(cluster.cluster_centers_))
print "Saved microstate assignments and microstate centers at %s" %out_path
return cluster.cluster_centers_, np.array(ev0), np.array(ev1)
def clustering(N_cluster_opt, dataset, traj):
cluster = KMeans(n_clusters=N_cluster_opt,
init='k-means++',
n_init=10,
max_iter=300,
tol=0.001,
precompute_distances='auto',
verbose=0,
random_state=None,
copy_x=True,
n_jobs=2).fit(dataset)
cluster_centers = cluster.cluster_centers_
print("center lenghth: " + str(len(cluster_centers)) + "\n")
clusters = [[] for i in range(0, N_cluster_opt)]
clusters_xyz = [[] for i in range(0, N_cluster_opt)]
clusters_xyz_center = []
fileout_labels = open(
"./AlleyCat-Ca-constrained/Labels_for_" + str(N_cluster_opt) +
"_clusters.dat", 'w')
for i in range(0, len(cluster.labels_[0])):
fileout_labels.write("snapshot " + str(i + 1) +
" corresponds to Cluster " +
str(cluster.labels_[0][i] + 1) + "\n")
for j in range(0, N_cluster_opt):
if cluster.labels_[0][i] == j:
clusters[j].append(dataset[0][i])
clusters_xyz[j].append(traj[i].xyz)
fileout = open(
"./AlleyCat-Ca-constrained/population_for_" + str(N_cluster_opt) +
"_clusters.dat", 'w')
for l in range(0, N_cluster_opt):
clusters_xyz_center.append(
np.average(np.array(clusters_xyz[l]), axis=0)[0])
fileout.write('The population of cluster ' + str(l) + ' is ' +
str(len(clusters[l])) + '\n')
print('The population of cluster ' + str(l) + ' is ' +
str(len(clusters[l])))
fileout_labels.close()
fileout.close()
return clusters_xyz, clusters_xyz_center, cluster_centers, clusters, cluster.labels_[
0]
n_samples = 200
topFile='NarK-strip.pdb'
dataset = []
ls = []
for i in sorted(glob.glob('*.npy')):
a = np.load(i)
b = np.array(a)
dataset.append(b)
ls.append(i)
print(i)
np.save('list', ls)
#trajs = [np.load('data.npy')]
# make cluster of the tICs trajectories
cluster = KMeans(n_clusters=myn_clusters)
cluster.fit(dataset)
l = cluster.labels_
T = []
for trj in glob.glob('*strip.mdcrd'):
T.append(trj)
T.sort()
# Write the output file, which have the information about population of each cluster,
# trajectory name and frame number of corresponding frame
asFunctions.writeOPF(l, T, myn_clusters, n_samples)
# Based on information in output file, build the cpptraj input file
asFunctions.CpptrajInGen_commonTop(topFile)
#pickle.dump( cluster , open( "tICCluster.pkl", "wb"))
f = DihedralFeaturizer(sincos=False)
dump(f, "raw_featurizer.pkl")
feat = f.transform(trj_list)
dump(feat, "raw_features.pkl")
f = load("./featurizer.pkl")
dump(f, "featurizer.pkl")
df1 = pd.DataFrame(f.describe_features(trj_list[0]))
dump(df1, "feature_descriptor.pkl")
feat = f.transform(trj_list)
dump(feat, "features.pkl")
t = tICA(lag_time=100, n_components=2, kinetic_mapping=False)
tica_feat = t.fit_transform(feat)
dump(t, "tica_mdl.pkl")
dump(tica_feat, "tica_features.pkl")
kmeans_mdl = KMeans(50)
ass = kmeans_mdl.fit_predict(tica_feat)
msm_mdl = MarkovStateModel(100)
msm_mdl.fit(ass)
dump(kmeans_mdl, "kmeans_mdl.pkl")
dump(ass, "assignments.pkl")
dump(msm_mdl, "msm_mdl.pkl")
def calculate_tica_components(self, cluster_method, calculate_strides=False, feats=None):
'''Load in the features, calculate a given number of tICA components (tica_components) given a
lagtime (lag_time), and save tICA coordinates and eigenvector data. It then creates and populates
a list for each desired component, clusters the data, saving normalized populations as populations.dat
and saving each cluster center as a .pdb. tICA plots are created and saved, and implied timescales are
calculated, saved, and plotted.
'''
# tICA parameters
tica_lagtime = 10 # determine from implied timescales
tica_components = 8 # how many tICs to compute
n_clusters = 100 # denotes number of microstates
n_timescales = tica_components # plot all eigenvalues --> timescales
md_time_step = 0.02 # ns
subsampled_time_step = 1. # ns multiplier of timescales and lagtimes in implied timescale plot
stride = int(subsampled_time_step / md_time_step) #time step stride for sub-sampling
equil_time = 1. # ns
equil_steps = 1 #int(equil_time / md_time_step) time steps to be removed from start
lagtimes = np.array([1,2,4,8,16,32,64,128,256,512,1024])
cluster_method = 'kcenters' # 'kcenters/kmeans'
all_ticas = list(itertools.permutations(range(1,tica_components+1), 2)) # all combinations
all_ticas = [[1,2]] # override: just show analysis for first two components
cluster_percentage_cutoff = 5 # clusters with a relative population less than this
# number will not be labeled on plot i.e. 0 : all clusters labeled
verbose = False
print("\nCalculating tICA components...")
# Load in feature files THIS WILL NEED TO BE CHANGED
if feats == None:
if calculate_strides:
self.calculate_stride_distances(stride, equil_steps)
data = np.load('/home/server/git/fah-scripts/DataAnalysisScripts/stride_dist/stride_dist_%d.npy' % self.proj_num)
else:
data = self.data
else:
data = np.load(feats)
features = []
for run in data:
for clone in run:
gen_seq = []
for gen in clone:
if gen is not None and gen[0] is not None:
if calculate_strides or feats is not None:
gen_seq.append(gen)
else:
gen_seq.append(gen[::stride])
if len(gen_seq) > 0:
gen_cat = np.concatenate(gen_seq)
if calculate_strides:
features.append(gen_cat)
else:
features.append(gen_cat[equil_steps:])
features = np.asarray(features)
print(features.shape)
print(features[0].shape)
tica_coordinates = tICA(lag_time=tica_lagtime,
n_components=int(tica_components)).fit_transform(features)
np.save('%s/lag_%d_coord_%d.npy' %(self.tICA_dir, tica_lagtime, tica_components), tica_coordinates)
# Initiate and populate an array for each component
for i in range(tica_components):
exec('tica_' + str(i+1) + ' = []')
for i in tqdm.tqdm(range(len(features))):
for j in range(len(tica_coordinates[i])):
for k in range(tica_components):
exec('tica_' + str(k+1) + '.append(tica_coordinates[i][j][k])')
# Perform clustering based on the cluster_method parameter.
if cluster_method == 'kcenters':
print("Clustering via KCenters...")
clusters = KCenters(n_clusters)
elif cluster_method == 'kmeans':
print("Clustering via KMeans...")
clusters = KMeans(n_clusters)
else:
sys.exit("Invalid cluster_method. Use kmeans or kcenters.")
# Determine cluster assignment for each frame.
sequences = clusters.fit_transform(tica_coordinates)
np.save('%s/lag_%d_clusters_%d_sequences.npy' %(self.tICA_dir, tica_lagtime, n_clusters), sequences)
np.save('%s/lag_%d_clusters_%d_center.npy' %(self.tICA_dir, tica_lagtime, n_clusters),
clusters.cluster_centers_)
# Determine cluster populations, normalize the counts, and save as percentages for
# labeling if a cluster contains more than cluster_percentage_cutoff percent of the data.
# Finally, save normalized counts.
print("\nDetermining cluster populations...")
if not os.path.exists('%s/cluster_centers' % self.tICA_dir):
os.makedirs('%s/cluster_centers' % self.tICA_dir)
counts = np.array([len(np.where(np.concatenate(sequences)==i)[0]) for i in range(n_clusters)])
normalized_counts = counts/float(counts.sum())
percentages = [ i*100 for i in normalized_counts ]
population_labels = [ [i,"%.2f"%percentages[i]] for i in range(len(percentages)) if percentages[i] > cluster_percentage_cutoff ]
np.savetxt('%s/cluster_centers/populations.dat' % self.tICA_dir, normalized_counts)
# Plot all unique combinations of tICA components
print("\nPlotting tICA components with cluster centers...")
all_ticas = list(itertools.permutations(range(1,tica_components+1), 2))
for j in tqdm.tqdm(range(len(all_ticas))): # For each pair
if all_ticas[j][0] < all_ticas[j][1]:
plt.figure(j, figsize=(20,16))
plt.hexbin(eval("tica_"+str(all_ticas[j][0])), eval("tica_"+str(all_ticas[j][1])), bins='log')
x_centers = [clusters.cluster_centers_[i][all_ticas[j][0]-1] for i in range(len(clusters.cluster_centers_))]
y_centers = [clusters.cluster_centers_[i][all_ticas[j][1]-1] for i in range(len(clusters.cluster_centers_))]
high_pop_x_centers = [ x_centers[i] for i in range(len(x_centers)) if percentages[i] > cluster_percentage_cutoff ]
high_pop_y_centers = [ y_centers[i] for i in range(len(y_centers)) if percentages[i] > cluster_percentage_cutoff ]
plt.plot(x_centers, y_centers, color='y', linestyle="", marker="o")
plt.plot(eval("tica_"+str(all_ticas[j][0])+'[0]'), eval("tica_"+str(all_ticas[j][1])+'[0]'), color='k', marker='*',markersize=24)
plt.xlabel('tic'+str(all_ticas[j][0]))
plt.ylabel('tic'+str(all_ticas[j][1]))
plt.title(self.proj_num)
# Add labels for high-population cluster centers
for label, x, y in zip(population_labels, high_pop_x_centers, high_pop_y_centers):
plt.annotate(
label,
xy = (x, y), xytext = (-15, 15),
textcoords = 'offset points', ha = 'right', va = 'bottom',
bbox = dict(boxstyle = 'round,pad=0.5', fc = 'yellow', alpha = 0.5),
arrowprops = dict(arrowstyle = '->', connectionstyle = 'arc3,rad=0'))
plt.savefig('%s/tica_' % (self.tICA_dir) +str(all_ticas[j][0])+'_'+str(all_ticas[j][1])+'.png')
plt.close()
###########################################################################
for filename in os.listdir(self.tICA_dir + '/cluster_centers'):
if filename.endswith('.pdb'):
os.remove(self.tICA_dir + '/cluster_centers/' + filename)
# Write out PDBs for each cluster center
print("Performing cluster analytics and saving center PDBs...\n")
runs, clones, gens = data.shape[0], data.shape[1], data.shape[2]
x, y, z = 0, 0, 0
for i in range(len(features)):
if i % clones == 0 and i != 0:
x += 1
if i % gens == 0:
y = 0
n_snapshots = len(clusters.distances_[i])
# Determine frames that are cluster centers
cluster_indices = np.arange(n_snapshots)[ (clusters.distances_[i] < 1e-6) ]
# Determine number of each cluster, correlates to populations.dat
cluster_labels = sequences[i][cluster_indices]
# Save each cluster center as a pdb
if list(cluster_indices): # load center-containing xtcs to check length
traj_cat = []
print('x: %d, y: %d, z: %d' % (x, y, z))
while True:
try:
traj = base_dir + 'PROJ%s/RUN%s/CLONE%s/results%s/traj_comp.xtc' % (self.proj_num, x, y, z)
traj_cat.append(md.load(traj, top=self.gro_file))
z += 1
except:
break
if len(traj_cat) > 0:
trajectory_file = md.join(traj_cat)
xtc_len = len(trajectory_file)
y += 1
z = 0
for j in range(len(cluster_indices)):
frames = range(xtc_len) # map the strided frame number back to xtc frame number
strided_frames = frames[equil_steps:][::stride]
xtc_frame = frames.index(strided_frames[cluster_indices[j]])
cluster_traj = trajectory_file[xtc_frame]
cluster_traj.save_pdb('%s/cluster_centers/state_%d_%.3f.pdb'%(self.tICA_dir, cluster_labels[j],percentages[cluster_labels[j]]))
if verbose:
print('Successfully saved PDB for cluster: %d, (rel.pop: %.3f)'%(cluster_labels[j],percentages[cluster_labels[j]]))
print('traj_file: %s (%d/%d)'%(trajectory_file,i,len(features)))
print('frame: %d (%d/%d centers from this trajectory)'%(cluster_indices[j],j,len(cluster_indices)))
print('strided: npy_frame/npy_len = %d/%d = %f'%(cluster_indices[j],n_snapshots,cluster_indices[j]/n_snapshots))
print('re-mapped: orig_frame/xtc_len = %d/%d = %f\n'%(xtc_frame,xtc_len,xtc_frame/xtc_len))
import numpy as np
from msmbuilder.cluster import KMeans, KCenters
import mdtraj.io as io
# 100 microstates
cluster = KMeans(
n_clusters=100,
n_jobs=-1,
verbose=0,
max_iter=100,
tol=0.0001,
)
dataset = []
for i in range(4):
a = io.loadh('../on_tica_l20_s1_%d.h5' % i)['arr_0']
a = a[:, 0:3] # using first 3 tICs
dataset.append(a)
print a.shape
for i in range(20):
a = io.loadh('../on_tica_l20_s2_%d.h5' % i)['arr_0']
a = a[:, 0:3]
dataset.append(a)
print a.shape
for i in range(20):
a = io.loadh('../on_tica_l20_s3_%d.h5' % i)['arr_0']
a = a[:, 0:3]
dataset.append(a)
print a.shape
for i in range(20):
a = io.loadh('../on_tica_l20_s4_%d.h5' % i)['arr_0']
from msmbuilder.decomposition import tICA
from msmbuilder.cluster import KMeans,KCenters,KMedoids
from msmbuilder.msm import MarkovStateModel
verbose = True
tica_data=np.load('../../ticas_n_8.npy')
reduced_data = []
for i in range(len(tica_data)):
reduced_data.append(tica_data[i][::100,:])
if verbose:
print "Clustering."
kmeans = KMeans(n_clusters=1200).fit(reduced_data)
Gen_fn = "Gens.npy"
np.save(Gen_fn,kmeans.cluster_centers_)
if verbose:
print "Wrote: %s"%Gen_fn
model_dir = "kmeans_model_n_1200"
if not os.path.exists(model_dir):
os.makedirs(model_dir)
model_fn = os.path.join(model_dir,'kmeans-combined.pkl')
joblib.dump(kmeans,model_fn)
if verbose:
print "Saved cluster model to %s"%model_fn
if verbose:
print "Assigning.."
assignments = kmeans.predict(tica_data)
if verbose:
print "%s not exists!" % tica_fn
continue
tica_data = np.load(tica_fn)
results = []
n_clusters = [100, 200, 400, 600, 800, 1000, 1200, 1500, 2000, 2500, 3000]
#n_clusters = [1200,1500,2000]
#n_clusters = [3500,3500,4000,4500,5000,6000]
lagtime = 50
for n in n_clusters:
kmeans = KMeans(n_clusters=n, n_jobs=-1)
print "Clustering data to %d clusters..." % n
for fold in range(nFolds):
train_data = []
test_data = []
for i in range(len(tica_data)):
cv = KFold(len(tica_data[i]), n_folds=nFolds)
for current_fold, (train_index, test_index) in enumerate(cv):
if current_fold == fold:
train_data.append(tica_data[i][train_index])
test_data.append(tica_data[i][test_index])
reduced_train_data = sub_sampling_data(train_data, stride=100)
kmeans.fit(reduced_train_data)
assignments_train = kmeans.predict(train_data)
assignments_test = kmeans.predict(test_data)
msm = MarkovStateModel(lag_time=lagtime)
def compute_tica_components():
'''Load in the features, calculate a given number of tICA components (tica_components) given a
lagtime (lag_time), and save tICA coordinates and eigenvector data. It then creates and populates
a list for each desired component, clusters the data, saving normalized populations as populations.dat
and saving each cluster center as a .pdb. tICA plots are created and saved, and implied timescales are
calculated, saved, and plotted.
'''
verbose = False
save_pdb = True
color_by = 'cluster'
if verbose:
print("\nCalculating tICA components...")
if not os.path.exists(project_title + '/tica_%d'%n_clusters):
os.mkdir(project_title + '/tica_%d'%n_clusters)
# load in feature files and determine indices of unbiased ensembles
feature_files = []
for i in range(runs):
run_files = sorted(glob.glob(/features/' + "P*R%d_*npy"%i))
feature_files += run_files
if i in unbiased_runs:
unbiased_indices = [len(feature_files) - len(run_files),len(feature_files)]
features = [np.load(x) for x in feature_files]
# perform tICA calculation and extract score / eigenvectors
tica_coordinates = tICA(lag_time=tica_lagtime,
n_components=int(n_components)).fit_transform(features)
tica_components = tICA(lag_time=tica_lagtime,
n_components=int(n_components)).fit(features)
eigenvectors = np.transpose(tica_components.eigenvectors_)
tica_score = tica_components.score(features)
np.save('%s/tica_%d/tica_coords-lag_%d-comp_%d.npy' %(
project_title, n_clusters, tica_lagtime, n_components), tica_coordinates)
np.save('%s/tica_%d/tica_comps-lag_%d-comp_%d.npy' %(
project_title, n_clusters, tica_lagtime, n_components), tica_components)
# Perform clustering based on the cluster_method parameter.
if verbose:
print('Clustering via %s'%cluster_method)
if cluster_method == 'kcenters':
clusters = KCenters(n_clusters)
elif cluster_method == 'kmeans':
clusters = KMeans(n_clusters)
elif cluster_method == 'kmedoids':
clusters = KMedoids(n_clusters)
else:
sys.exit('Invalid cluster_method. Use kcenters/kmeans/kmedoids.')
# Cluster unbiased data and fit biased data to these centers
new_assignments = []
sequences = clusters.fit_transform(tica_coordinates[unbiased_indices[0]:unbiased_indices[1]])
for i in tqdm.tqdm_notebook(range(unbiased_indices[0])):
tica_traj = tica_coordinates[i]
if isinstance(tica_traj, np.ndarray):
if not (tica_traj.dtype == 'float32' or tica_traj.dtype == 'float64'):
tica_traj = tica_traj.astype('float64')
labels, inertia = msmbuilder.libdistance.assign_nearest(
tica_traj, clusters.cluster_centers_, metric='euclidean')
new_assignments.append(labels)
new_assignments += sequences # tack the unbiased assignments back on to the end.
np.save('%s/tica_%d/lag_%d_clusters_%d_assignments.npy' %(
project_title, n_clusters, tica_lagtime, n_clusters), new_assignments)
np.save('%s/tica_%d/lag_%d_clusters_%d_center.npy' %(
project_title, n_clusters, tica_lagtime, n_clusters), clusters.cluster_centers_)
# Determine cluster populations, normalize the counts, and save as percentages for
# labeling if a cluster contains more than cluster_percentage_cutoff percent of the data.
# Finally, save normalized counts.
if verbose:
print("\nDetermining cluster populations...")
if not os.path.exists('%s/tica_%d/%s_clusters'%(project_title,n_clusters,cluster_method)):
os.mkdir('%s/tica_%d/%s_clusters'%(project_title,n_clusters,cluster_method))
if not os.path.exists('%s/tica_%d/plots'%(project_title,n_clusters)):
os.mkdir('%s/tica_%d/plots'%(project_title,n_clusters))
counts = np.array([len(np.where(np.concatenate(sequences)==i)[0]) for i in range(n_clusters)])
normalized_counts = counts/float(counts.sum())
percentages = [ i*100 for i in normalized_counts ]
population_labels = [ [i,"%.2f"%percentages[i]] for i in range(len(percentages)) if percentages[i] > cluster_percentage_cutoff ]
np.savetxt('%s/tica_%d/%s_clusters/populations.dat'
%(project_title,n_clusters,cluster_method), normalized_counts)
# Plot all unique combinations of tICA components
if verbose:
print("\nPlotting tICA components...")
tica_coordinates = np.concatenate(tica_coordinates)
new_assignments = np.concatenate(new_assignments)
cluster_colors = matplotlib.cm.rainbow(np.linspace(0,1,n_clusters))
for j in tqdm.tqdm_notebook(range(len(all_ticas)),leave=False): # For each pair
if all_ticas[j][0] < all_ticas[j][1]:
plt.figure(j, figsize=(20,16))
tICx, tICy = all_ticas[j][0]-1, all_ticas[j][1]-1
plt.hexbin(tica_coordinates[:,tICx],tica_coordinates[:,tICy], bins='log')
for l in tqdm.tqdm(range(len(tica_coordinates))[::stride*2]):
if color_by == 'cluster':
plt.plot(tica_coordinates[l][tICx], tica_coordinates[l][tICy],
color=cluster_colors[new_assignments[l]], linestyle="", marker="o")
x_centers = [clusters.cluster_centers_[i][tICx] for i in range(len(clusters.cluster_centers_))]
y_centers = [clusters.cluster_centers_[i][tICy] for i in range(len(clusters.cluster_centers_))]
high_pop_x_centers = [ x_centers[i] for i in range(len(x_centers)) if percentages[i] > cluster_percentage_cutoff ]
high_pop_y_centers = [ y_centers[i] for i in range(len(y_centers)) if percentages[i] > cluster_percentage_cutoff ]
plt.plot(x_centers, y_centers, color='y', linestyle="", marker="o")
plt.plot(tica_coordinates[:,tICx][0],tica_coordinates[:,tICy][0], color='k', marker='*',markersize=24)
plt.xlabel('tIC'+str(all_ticas[j][0]))
plt.ylabel('tIC'+str(all_ticas[j][1]))
plt.title(project_title)
# Add labels for high-population cluster centers
for label, x, y in zip(population_labels, high_pop_x_centers, high_pop_y_centers):
plt.annotate(
label,
xy = (x, y), xytext = (-15, 15),
textcoords = 'offset points', ha = 'right', va = 'bottom',
bbox = dict(boxstyle = 'round,pad=0.5', fc = 'yellow', alpha = 0.5),
arrowprops = dict(arrowstyle = '->', connectionstyle = 'arc3,rad=0'))
plt.savefig('%s/tica_%d/plots/tica_%d_%d.png'%(project_title,n_clusters,
all_ticas[j][0], all_ticas[j][1]))
plt.close()
# Write out PDBs for each cluster center
if verbose:
print("Performing cluster analytics and saving center PDBs...\n")
if save_pdb:
trajectory_files, feature_files, cluster_features = [],[],[]
for run in range(runs): # get only xtc files that correlate to cluster-center features
trajectory_files += [re.sub('features',
'traj_data/RUN%d'%run,re.sub('npy','xtc',x)
) for x in sorted(glob.glob('%s/features/*R%d_*npy'%(
project_title,run)))]
feature_files += sorted(glob.glob('%s/features/*R%d_*npy'%(project_title,run)))
for i in tqdm.tqdm_notebook(range(len(trajectory_files)),leave=False):
n_snapshots = len(clusters.distances_[i])
# Determine frames that are cluster centers
cluster_indices = np.arange(n_snapshots)[ (clusters.distances_[i] < 1e-6) ]
# Determine number of each cluster, correlates to populations.dat
cluster_labels = sequences[i][cluster_indices]
# Save each cluster center as a pdb
if list(cluster_indices): # load center-containing xtcs to check length
xtc_len = len(md.load(trajectory_files[i],top=structure_file))
# map strided frame number back to xtc frame number
for j in range(len(cluster_indices)):
frames = range(xtc_len)
strided_frames = frames[equil_steps:][::stride]
xtc_frame = frames.index(strided_frames[cluster_indices[j]])
cluster_traj = md.load_frame(trajectory_files[i], xtc_frame,
top=structure_file)
cluster_features.append(np.load(feature_files[i])[cluster_indices[j]])
cluster_traj.save_pdb('%s/tica_%d/%s_clusters/state_%d.pdb'
%(project_title,n_clusters,cluster_method,
cluster_labels[j]))
# save cluster information
with open('%s/tica_%d/cluster.dat'%(project_title,n_clusters),'w') as f:
f.write('\nSuccessfully saved PDB for cluster: %d, (rel.pop: %.3f)'%(
cluster_labels[j],percentages[cluster_labels[j]]))
f.write('traj_file: %s (%d/%d)'%(trajectory_files[i],i,len(features)))
f.write('frame: %d (%d/%d centers from this trajectory)'%(
cluster_indices[j],j,len(cluster_indices)))
f.write('strided: npy_frame/npy_len = %d/%d = %f'%(
cluster_indices[j],n_snapshots,cluster_indices[j]/n_snapshots))
f.write('re-mapped: orig_frame/xtc_len = %d/%d = %f\n'%(
xtc_frame,xtc_len,xtc_frame/xtc_len))
f.close()
# save features corresponding to each cluster center
np.save('%s/tica_%d/cluster_features.npy'%(project_title,n_clusters),cluster_features)
return tica_score