notebook.community

Edit and run 





In [5]:



    


%matplotlib inline



    











In [50]:



    


import sys
import numpy as np
import pandas as pd
import seaborn as sns
import statsmodels
import sklearn
import matplotlib.pyplot as plt
sys.path.append('PATH2MIMOSCA_CODE')
import mimosca as mimosca



    











In [10]:



    


DGE1=mimosca.read_10x('PATH2MM10_CHANGETHIS/mm10/')
DGE_genes=mimosca.collapse2gene(DGE1)



    











In [11]:



    


DGE_genes.head()



    


















    
Out[11]:










  
    

      
      AAACATTGACTACG
      AAACGCACCGTTAG
      AAACGCTGACGGTT
      AAACGCTGTCTCGC
      AAACGGCTCGAATC
      AAAGATCTACCCAA
      AAAGCCTGAACCTG
      AAAGCCTGCTCTTA
      AAAGCCTGTCACGA
      AAAGTTTGTTCTAC
      ...
      TTGTCATGTCTCGC
      TTTAGCTGAGCAAA
      TTTAGCTGCATTCT
      TTTCACGACACCAA
      TTTCACGACTCTCG
      TTTCAGTGAGCAAA
      TTTCAGTGGCGATT
      TTTCAGTGTGCAGT
      TTTCCAGAATGGTC
      TTTCTACTTGACCA
    

  
  
    

      0610007P14Rik
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      ...
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      1.0
      0.0
      0.0
    

    

      0610009B22Rik
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      ...
      1.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
    

    

      0610009L18Rik
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      ...
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
    

    

      0610009O20Rik
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      ...
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
    

    

      0610010F05Rik
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      ...
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
      0.0
    

  



5 rows × 1310 columns



















In [36]:



    


#select variable genes
highlight=['Tnf','Il6','Rsad2','Cxcl10']
var_genes=mimosca.fano_variable(mimosca.tp10k_transform(DGE_genes),input_mean=np.log2(DGE_genes.mean(axis=1)+1),resthresh=0.02,meanthresh=0.5,highlight_genes=highlight,plot=1)
these_genes=var_genes[np.logical_or(var_genes['residuals']>0.01,var_genes['mean']>2)].index



    


















    
























In [37]:



    


# Subset to variable genes
DGE_wt = DGE_genes.loc[these_genes]



    











In [38]:



    


# Z normalize Genes
DGEwtZ = mimosca.Zgenes_floor(np.log2(mimosca.tp10k_transform(DGE_wt)+1),floor=0.1)#,batchvec=batchwt)



    











In [39]:



    


#Facebook's Fast PCA (useful if using more than 10,000 cells) for calculating SVD, top 50 components
[Ufb,Sfb,Vfb]=mimosca.fb_pca(DGEwtZ,k=50)



    











In [40]:



    


#Genes detected per cell
cqc=(DGE_wt>0).sum()



    











In [41]:



    


#Calculate correlation of PC scores with genes detected
qc_cor=[]
for pc in Vfb.columns:
    qc_cor.append(np.corrcoef(Vfb[pc],cqc)[0][1])
plt.plot(qc_cor)
plt.xlabel('PCs')
plt.ylabel('Correlation with Genes Detected')



    


















    
Out[41]:








<matplotlib.text.Text at 0x125fb59d0>

The Permutation and Noise Envelope Calculations Below Give The User Some Sense of the Number of PCs that are "significant"





In [18]:



    


# Permute gene expression matrix and calculate SVD
DGE_perm=mimosca.permute_matrix(DGE_wt)
DGEwtZperm = mimosca.Zgenes(np.log2(mimosca.tp10k_transform(DGE_perm)+1))
[Ufbperm,Sfbperm,Vfbperm]=mimosca.fb_pca(DGEwtZperm,k=50)
plt.plot(Sfb,label='Actual Data')
plt.plot(Sfbperm,label='Permuted Data')
plt.legend()
plt.ylabel('Singular Values')
plt.xlabel('Ranked Eigenvectors')



    


















    
Out[18]:








<matplotlib.text.Text at 0x1108bfe10>










    
























In [19]:



    


#calculate sensitivity of various PCs to noise and show noise envelope profile
PC_cor=mimosca.PC_noise(DGEwtZ)

for i in range(20):
    plt.plot(np.linspace(-2,2,20),PC_cor.iloc[i],label=i)
plt.legend()
plt.xlabel('Log10(Added Noise)')
plt.ylabel('Correlation with Original Eigenvector')



    


















    
Out[19]:








<matplotlib.text.Text at 0x112f82810>










    
























In [42]:



    


PVALS=mimosca.jackstraw(DGEwtZ,sig_pcs=10,reps=1000,per=0.005)



    











In [43]:



    


#Example of Jackstraw p-value calculation
plt.scatter(Ufb[0],PVALS[0])
plt.axhline(1.3,c='red')
plt.axhline(-1.3,c='red')
plt.ylabel('Signed Log10(q-value) Jackstraw')
plt.xlabel('PCA loadings')
plt.title('PC2')



    


















    
Out[43]:








<matplotlib.text.Text at 0x112dd5410>










    
























In [120]:



    


df_pc = mimosca.PCA2GO(DGEwtZ,thresh=1.3,repin=1000,species='mouse')



    


















    






done with jackstraw
load obo file /Users/atray/Desktop/SANDBOX/trial_scripts/go_analysis/db/go-basic.obo
/Users/atray/Desktop/SANDBOX/trial_scripts/go_analysis/db/go-basic.obo: fmt(1.2) rel(2016-06-04) 46,686 GO Terms
19,070 annotated genes
   296 out of    340 population items found in association
0
Calculating uncorrected p-values using fisher
    23 out of     23 study items found in association
Running multitest correction: statsmodels fdr_bh
  512 GO terms are associated with 23 of 23 study items
  2,302 GO terms are associated with 296 of 340 population items
Calculating uncorrected p-values using fisher
    21 out of     23 study items found in association
Running multitest correction: statsmodels fdr_bh
  273 GO terms are associated with 21 of 23 study items
  2,302 GO terms are associated with 296 of 340 population items
1
Calculating uncorrected p-values using fisher
    20 out of     26 study items found in association
Running multitest correction: statsmodels fdr_bh
  299 GO terms are associated with 20 of 26 study items
  2,302 GO terms are associated with 296 of 340 population items
Calculating uncorrected p-values using fisher
    21 out of     21 study items found in association
Running multitest correction: statsmodels fdr_bh
  239 GO terms are associated with 21 of 21 study items
  2,302 GO terms are associated with 296 of 340 population items
2
Calculating uncorrected p-values using fisher
    25 out of     25 study items found in association
Running multitest correction: statsmodels fdr_bh
  285 GO terms are associated with 25 of 25 study items
  2,302 GO terms are associated with 296 of 340 population items
Calculating uncorrected p-values using fisher
    22 out of     31 study items found in association
Running multitest correction: statsmodels fdr_bh
  246 GO terms are associated with 22 of 31 study items
  2,302 GO terms are associated with 296 of 340 population items
3
Calculating uncorrected p-values using fisher
    21 out of     21 study items found in association
Running multitest correction: statsmodels fdr_bh
  458 GO terms are associated with 21 of 21 study items
  2,302 GO terms are associated with 296 of 340 population items
Calculating uncorrected p-values using fisher
    22 out of     26 study items found in association
Running multitest correction: statsmodels fdr_bh
  363 GO terms are associated with 22 of 26 study items
  2,302 GO terms are associated with 296 of 340 population items
4
Calculating uncorrected p-values using fisher
    21 out of     25 study items found in association
Running multitest correction: statsmodels fdr_bh
  325 GO terms are associated with 21 of 25 study items
  2,302 GO terms are associated with 296 of 340 population items
Calculating uncorrected p-values using fisher
    26 out of     27 study items found in association
Running multitest correction: statsmodels fdr_bh
  256 GO terms are associated with 26 of 27 study items
  2,302 GO terms are associated with 296 of 340 population items
5
Calculating uncorrected p-values using fisher
    19 out of     20 study items found in association
Running multitest correction: statsmodels fdr_bh
  311 GO terms are associated with 19 of 20 study items
  2,302 GO terms are associated with 296 of 340 population items
Calculating uncorrected p-values using fisher
    22 out of     23 study items found in association
Running multitest correction: statsmodels fdr_bh
  295 GO terms are associated with 22 of 23 study items
  2,302 GO terms are associated with 296 of 340 population items
6
Calculating uncorrected p-values using fisher
    21 out of     22 study items found in association
Running multitest correction: statsmodels fdr_bh
  242 GO terms are associated with 21 of 22 study items
  2,302 GO terms are associated with 296 of 340 population items
Calculating uncorrected p-values using fisher
    21 out of     21 study items found in association
Running multitest correction: statsmodels fdr_bh
  435 GO terms are associated with 21 of 21 study items
  2,302 GO terms are associated with 296 of 340 population items
7
Calculating uncorrected p-values using fisher
    27 out of     28 study items found in association
Running multitest correction: statsmodels fdr_bh
  393 GO terms are associated with 27 of 28 study items
  2,302 GO terms are associated with 296 of 340 population items
Calculating uncorrected p-values using fisher
    22 out of     23 study items found in association
Running multitest correction: statsmodels fdr_bh
  233 GO terms are associated with 22 of 23 study items
  2,302 GO terms are associated with 296 of 340 population items
8
Calculating uncorrected p-values using fisher
    22 out of     23 study items found in association
Running multitest correction: statsmodels fdr_bh
  375 GO terms are associated with 22 of 23 study items
  2,302 GO terms are associated with 296 of 340 population items
Calculating uncorrected p-values using fisher
    22 out of     25 study items found in association
Running multitest correction: statsmodels fdr_bh
  328 GO terms are associated with 22 of 25 study items
  2,302 GO terms are associated with 296 of 340 population items
9
Calculating uncorrected p-values using fisher
    22 out of     23 study items found in association
Running multitest correction: statsmodels fdr_bh
  479 GO terms are associated with 22 of 23 study items
  2,302 GO terms are associated with 296 of 340 population items
Calculating uncorrected p-values using fisher
    21 out of     25 study items found in association
Running multitest correction: statsmodels fdr_bh
  263 GO terms are associated with 21 of 25 study items
  2,302 GO terms are associated with 296 of 340 population items
10
Calculating uncorrected p-values using fisher
    18 out of     20 study items found in association
Running multitest correction: statsmodels fdr_bh
  253 GO terms are associated with 18 of 20 study items
  2,302 GO terms are associated with 296 of 340 population items
Calculating uncorrected p-values using fisher
    21 out of     23 study items found in association
Running multitest correction: statsmodels fdr_bh
  267 GO terms are associated with 21 of 23 study items
  2,302 GO terms are associated with 296 of 340 population items
11
Calculating uncorrected p-values using fisher
    21 out of     24 study items found in association
Running multitest correction: statsmodels fdr_bh
  406 GO terms are associated with 21 of 24 study items
  2,302 GO terms are associated with 296 of 340 population items
Calculating uncorrected p-values using fisher
    24 out of     26 study items found in association
Running multitest correction: statsmodels fdr_bh
  279 GO terms are associated with 24 of 26 study items
  2,302 GO terms are associated with 296 of 340 population items
12
Calculating uncorrected p-values using fisher
    21 out of     30 study items found in association
Running multitest correction: statsmodels fdr_bh
  235 GO terms are associated with 21 of 30 study items
  2,302 GO terms are associated with 296 of 340 population items
Calculating uncorrected p-values using fisher
    19 out of     24 study items found in association
Running multitest correction: statsmodels fdr_bh
  327 GO terms are associated with 19 of 24 study items
  2,302 GO terms are associated with 296 of 340 population items
13
Calculating uncorrected p-values using fisher
    22 out of     24 study items found in association
Running multitest correction: statsmodels fdr_bh
  331 GO terms are associated with 22 of 24 study items
  2,302 GO terms are associated with 296 of 340 population items
Calculating uncorrected p-values using fisher
    27 out of     31 study items found in association
Running multitest correction: statsmodels fdr_bh
  326 GO terms are associated with 27 of 31 study items
  2,302 GO terms are associated with 296 of 340 population items
14
Calculating uncorrected p-values using fisher
    18 out of     25 study items found in association
Running multitest correction: statsmodels fdr_bh
  312 GO terms are associated with 18 of 25 study items
  2,302 GO terms are associated with 296 of 340 population items
Calculating uncorrected p-values using fisher
    24 out of     35 study items found in association
Running multitest correction: statsmodels fdr_bh
  298 GO terms are associated with 24 of 35 study items
  2,302 GO terms are associated with 296 of 340 population items

















In [129]:



    


cg=sns.clustermap(df_pc,figsize=(10,20))
a=plt.setp(cg.ax_heatmap.yaxis.get_majorticklabels(),rotation=0)



    


















    
























In [101]:



    


# Select 9 PCs based on the qc plots above
Vfb_cluster=Vfb.ix[:,range(9)]



    











In [102]:



    


model = sklearn.manifold.TSNE(n_components=2, perplexity=20,verbose=2,init='pca',n_iter_without_progress=10000,min_grad_norm=0)
T_sne=model.fit_transform(Vfb_cluster)



    


















    






[t-SNE] Computing pairwise distances...
[t-SNE] Computing 61 nearest neighbors...
[t-SNE] Computed conditional probabilities for sample 1000 / 1310
[t-SNE] Computed conditional probabilities for sample 1310 / 1310
[t-SNE] Mean sigma: 0.021790
[t-SNE] Iteration 25: error = 1.8454133, gradient norm = 0.0098836
[t-SNE] Iteration 50: error = 1.7922612, gradient norm = 0.0110613
[t-SNE] Iteration 75: error = 1.4431548, gradient norm = 0.0042804
[t-SNE] Iteration 100: error = 1.4159998, gradient norm = 0.0073687
[t-SNE] Error after 100 iterations with early exaggeration: 1.416000
[t-SNE] Iteration 125: error = 1.3150527, gradient norm = 0.0024946
[t-SNE] Iteration 150: error = 1.3109168, gradient norm = 0.0057507
[t-SNE] Iteration 175: error = 1.3300472, gradient norm = 0.0070111
[t-SNE] Iteration 200: error = 1.3546096, gradient norm = 0.0073184
[t-SNE] Iteration 200: did not make any progress during the last 30 episodes. Finished.
[t-SNE] Error after 200 iterations: 1.354610

















In [103]:



    


#Tsne plot colored by genes detected
T_sne=pd.DataFrame(T_sne)
plt.scatter(T_sne[0],T_sne[1],alpha=0.75,c=cqc,cmap='bwr')



    


















    
Out[103]:








<matplotlib.collections.PathCollection at 0x139defed0>










    
























In [104]:



    


# You can use infomap if you have the libraries installed, this is probably slightly overclustered
info_labels=['cluster-'+str(x) for x in sklearn.cluster.k_means(Vfb_cluster,7)[1]]



    











In [105]:



    


info_labels=np.array(info_labels)
unique_labs=np.unique(info_labels)
len(unique_labs)



    


















    
Out[105]:








7

















In [106]:



    


Tsneplot=T_sne.copy()
Tsneplot.columns=['1','2']
Tsneplot['clustering']=np.array(info_labels)
sns.lmplot('1','2',data=Tsneplot,hue='clustering',fit_reg=False)



    


















    
Out[106]:








<seaborn.axisgrid.FacetGrid at 0x12f712450>










    
























In [107]:



    


DGE_wtall=DGE1



    











In [108]:



    


DGE_Z_expressed=mimosca.Zgenes(np.log2(mimosca.tp10k_transform(DGE_wtall[DGE_wtall.sum(axis=1)>20])+1))



    











In [109]:



    


#calculate pairwsie differential expression between each cluster and if below some threshold merge
df_DE=mimosca.cluster_merger(DGE_Z_expressed,Vfb_cluster.index,info_labels)
df_DE



    


















    






0
1
2
3
4
5
6










    
Out[109]:










  
    

      
      0
      1
      2
      3
      4
      5
      6
    

  
  
    

      0
      0.0
      282.0
      953.0
      509.0
      544.0
      961.0
      385.0
    

    

      1
      282.0
      0.0
      1226.0
      318.0
      607.0
      1023.0
      147.0
    

    

      2
      953.0
      1226.0
      0.0
      1927.0
      881.0
      1153.0
      1828.0
    

    

      3
      509.0
      318.0
      1927.0
      0.0
      779.0
      1347.0
      162.0
    

    

      4
      544.0
      607.0
      881.0
      779.0
      0.0
      332.0
      701.0
    

    

      5
      961.0
      1023.0
      1153.0
      1347.0
      332.0
      0.0
      1252.0
    

    

      6
      385.0
      147.0
      1828.0
      162.0
      701.0
      1252.0
      0.0
    

  




















In [110]:



    


#differential expression per cluster, each cluster against all others
df_minp = pd.DataFrame()
allcells=set(Vfb_cluster.index)
df_DC_cluster=pd.DataFrame()

for lab1 in unique_labs:
    cells1=Vfb_cluster.index[info_labels==lab1]
    cells2=list(allcells-set(cells1))
    print(lab1,len(cells1))

    fc,sl10pval=mimosca.ttest_DGE(DGE_Z_expressed[cells1],DGE_Z_expressed[cells2])
    #sl10pval=np.array(pd.DataFrame(sl10pval))
    df_minp[lab1]=np.sign(sl10pval)*(-np.log10(statsmodels.sandbox.stats.multicomp.fdrcorrection0(np.power(10,-np.abs(sl10pval)))[1]))#sl10pval



    


















    






('cluster-0', 254)
('cluster-1', 217)
('cluster-2', 108)
('cluster-3', 242)
('cluster-4', 84)
('cluster-5', 112)
('cluster-6', 293)

















In [111]:



    


#convert infinity values to finite numbers
df_minp.index=DGE_Z_expressed.index
df_minp[df_minp>(1000.0)]=1000.0
df_minp[df_minp<(-1000.0)]=-1000.0



    











In [112]:



    


df_minp.index=mimosca.genenames_from10x(df_minp.index)



    











In [113]:



    


df_DEGO=mimosca.DE2GO(df_minp,df_minp.index,sig_thresh=1.3,species='mouse')



    


















    






load obo file /Users/atray/Desktop/SANDBOX/trial_scripts/go_analysis/db/go-basic.obo
/Users/atray/Desktop/SANDBOX/trial_scripts/go_analysis/db/go-basic.obo: fmt(1.2) rel(2016-06-04) 46,686 GO Terms
19,070 annotated genes
 4,746 out of  5,197 population items found in association
cluster-0
Calculating uncorrected p-values using fisher
    82 out of    100 study items found in association
Running multitest correction: statsmodels fdr_bh
  994 GO terms are associated with 82 of 100 study items
  10,678 GO terms are associated with 4,746 of 5,197 population items
Calculating uncorrected p-values using fisher
 1,224 out of  1,310 study items found in association
Running multitest correction: statsmodels fdr_bh
  5,679 GO terms are associated with 1,224 of 1,310 study items
  10,678 GO terms are associated with 4,746 of 5,197 population items
cluster-1
Calculating uncorrected p-values using fisher
   255 out of    292 study items found in association
Running multitest correction: statsmodels fdr_bh
  2,135 GO terms are associated with 255 of 292 study items
  10,678 GO terms are associated with 4,746 of 5,197 population items
Calculating uncorrected p-values using fisher
   394 out of    432 study items found in association
Running multitest correction: statsmodels fdr_bh
  2,786 GO terms are associated with 394 of 432 study items
  10,678 GO terms are associated with 4,746 of 5,197 population items
cluster-2
Calculating uncorrected p-values using fisher
   151 out of    167 study items found in association
Running multitest correction: statsmodels fdr_bh
  1,464 GO terms are associated with 151 of 167 study items
  10,678 GO terms are associated with 4,746 of 5,197 population items
Calculating uncorrected p-values using fisher
 3,254 out of  3,574 study items found in association
Running multitest correction: statsmodels fdr_bh
  9,042 GO terms are associated with 3,254 of 3,574 study items
  10,678 GO terms are associated with 4,746 of 5,197 population items
cluster-3
Calculating uncorrected p-values using fisher
   875 out of    947 study items found in association
Running multitest correction: statsmodels fdr_bh
  4,883 GO terms are associated with 875 of 947 study items
  10,678 GO terms are associated with 4,746 of 5,197 population items
Calculating uncorrected p-values using fisher
   246 out of    291 study items found in association
Running multitest correction: statsmodels fdr_bh
  1,968 GO terms are associated with 246 of 291 study items
  10,678 GO terms are associated with 4,746 of 5,197 population items
cluster-4
Calculating uncorrected p-values using fisher
   393 out of    463 study items found in association
Running multitest correction: statsmodels fdr_bh
  2,752 GO terms are associated with 393 of 463 study items
  10,678 GO terms are associated with 4,746 of 5,197 population items
Calculating uncorrected p-values using fisher
   839 out of    898 study items found in association
Running multitest correction: statsmodels fdr_bh
  5,126 GO terms are associated with 839 of 898 study items
  10,678 GO terms are associated with 4,746 of 5,197 population items
cluster-5
Calculating uncorrected p-values using fisher
   660 out of    738 study items found in association
Running multitest correction: statsmodels fdr_bh
  3,678 GO terms are associated with 660 of 738 study items
  10,678 GO terms are associated with 4,746 of 5,197 population items
Calculating uncorrected p-values using fisher
 1,541 out of  1,647 study items found in association
Running multitest correction: statsmodels fdr_bh
  6,633 GO terms are associated with 1,541 of 1,647 study items
  10,678 GO terms are associated with 4,746 of 5,197 population items
cluster-6
Calculating uncorrected p-values using fisher
   469 out of    510 study items found in association
Running multitest correction: statsmodels fdr_bh
  3,361 GO terms are associated with 469 of 510 study items
  10,678 GO terms are associated with 4,746 of 5,197 population items
Calculating uncorrected p-values using fisher
   257 out of    289 study items found in association
Running multitest correction: statsmodels fdr_bh
  1,992 GO terms are associated with 257 of 289 study items
  10,678 GO terms are associated with 4,746 of 5,197 population items

















In [127]:



    


#plot GO enrichments (remove .sample(100) and asjust figsize if you want to see them all)
cg=sns.clustermap(df_DEGO.sample(100),figsize=(20,20),metric='correlation')
a=plt.setp(cg.ax_heatmap.yaxis.get_majorticklabels(),rotation=0)



    


















    
























In [115]:



    


Vfb_cluster_reproject=mimosca.project_ontoPC(Ufb.ix[:,range(9)],DGEwtZ)
from sklearn.ensemble import RandomForestClassifier
clf = RandomForestClassifier(n_estimators=100,n_jobs=-1,oob_score=True,class_weight='balanced')
clf = clf.fit(Vfb_cluster_reproject,info_labels)
OOB=pd.DataFrame(clf.oob_decision_function_)



    











In [128]:



    


#quantify how many cells are confidently classified
np.mean(OOB.max(axis=1)>0.5)



    


















    
Out[128]:








0.88473282442748091

















In [117]:



    


#plot AUC curves for class membership

clusterun=np.unique(info_labels)

for i in range(np.shape(OOB)[1]):
    [fpr,tpr,thresholds]=sklearn.metrics.roc_curve(info_labels==clusterun[i],OOB[i])
    plt.plot(fpr,tpr)
    plt.xlabel('False Positive Rate')
    plt.ylabel('True Positive Rate')



    


















    
























In [118]:



    


#Pseudo code for classifying new dataset

#DGEZ=New dataset you want to classify
#Vfb_new=mimosca.project_ontoPC(Ufb,DGEZ)
#PROBS=pd.DataFrame(clf.predict_proba(Vfb_new.ix[:,range(9)]))
#np.mean(PROBS.max(axis=1)>0.5)

Content source: asncd/MIMOSCA

Similar notebooks:

notebook.community | gallery | about