Identifying the biologically relevant gene categories based on gene expression and biological data: an example on prostate cancer

D. Huang, T. W. S. Chow 


Comparison of different vs

MI based forward searching process for hyper-gene selection

Evaluation framework based on classification

The prostate-cancer-related GO biological process terms

The prostate-cancer-related KEGG pathways

 

(PDF version

(Source code (in Matlab))


Comparisons of different vs 

Given a gene (say, gk) and a gene functional category (say ci), we estimate p(gk|ci) according to

.

s(gk, ci) is the similarity between gk and ci, and is determined by  where skj is the association between gk and gj and is measured using Pearsonís correlation. In our investigation, different settings of v were tested on the synthetic and real data. In table S1 and S2, the comparative results are presented.

 Table S1. The results on a synthetic example. A percentage value indicates the probability of the corresponding category being selected across 1000 trials.

 

(a) The results on the category setting shown in Fig. 2

v = 2

{S2, S5}: 80.1%; {S3, S5}: 2.3%; {S4, S5}: 0.4%.

v = 6

{S2, S5}: 89.1%; {S3, S5}: 1.2%; {S4, S5}: 0.

v = 10

{S2, S5}: 37.5%; {S3, S5}: 0; {S4, S5}: 0.

v =  18

{S2, S5}: 20.7%; {S3, S5}: 0; {S4, S5}: 0.

 

(b) The results on the category setting shown in Fig. 2

v = 2

{S2, S5}: 72%; {S3, S5}: 1.2%; {S4, S5}: 0.3%.

v = 6

{S2, S5}: 80.1%; {S3, S5}: 2.1%; {S4, S5}: 0.4%.

v = 10

{S2, S5}: 45.3%; {S3, S5}: 0.5%; {S4, S5}: 2.1%.

v =  18

{S2, S5}: 8.1%; {S3, S5}: 0; {S4, S5}: 0.

 

Back to Top

MI based forward searching process for hyper-gene selection 

This process can be stated as follows.

Step 1. Set the selected gene set, say S, with Empty.

Step 2. For each hyper-gene, say g, compute MI(g, y) where y is the response.

Step 3. Find the hyper-gene having the maximal MI(g, y), and place into S.

Step 4. Repeat the followings until certain hyper-genes have been selected.

(a) For each unselected hyper-gene, say g, calculate  MI(S+g, y).

(b)  Identify the one with  the maximal MI(S+g, y), and place that hyper-gene into S.

Step 5. Output the set S.

 

In the above process, 

,

where

,

,

.

Furthermore,  we have 

,

X is a given dataset. M and N are the dimensionality and size of X.

 

Back to Top

 


Evaluation framework based on classification

We evaluated the quality of a selected category set based on its classification capability. In this course, the framework of 5 cross validation was adopted. In Fig S1, this framework is detailed.

 

Fig S1 the block diagram of 5CV framework for classification evaluation

Back to Top

The prostate-cancer-related GO biological terms

Based on the prostate-cancer-related gene list (http://www.superarray.com/gene_array_product/HTML/OHS-403.html), we identify 58 biological process GO terms that are significantly prostate-cancer-related. 

GO:0016049:cell growth                                                                

GO:0048732:gland development                                                          

GO:0045893:positive regulation of transcription, DNA-dependent                        

GO:0030521:androgen receptor signaling pathway                                        

GO:0030518:steroid hormone receptor signaling pathway                                 

GO:0016567:protein ubiquitination                                                      

GO:0008637:apoptotic mitochondrial changes                                            

GO:0045884:regulation of survival gene product activity                               

GO:0007281:germ cell development                                                      

GO:0008634:negative regulation of survival gene product activity                      

GO:0008624:induction of apoptosis by extracellular signals                            

GO:0006916:anti-apoptosis                                                             

GO:0008629:induction of apoptosis by intracellular signals                            

GO:0000079:regulation of cyclin dependent protein kinase activity                     

GO:0007050:cell cycle arrest                                                          

GO:0001501:skeletal development                                                       

GO:0006935:chemotaxis                                                                 

GO:0006007:glucose catabolism                                                         

GO:0007169:transmembrane receptor protein tyrosine kinase signaling pathway           

GO:0008286:insulin receptor signaling pathway                                          

GO:0051262:protein tetramerization                                                    

GO:0008284:positive regulation of cell proliferation                                  

GO:0016064:humoral defense mechanism (sensu Vertebrata)                                

GO:0045793:positive regulation of cell size                                           

GO:0001558:regulation of cell growth                                                  

GO:0019735:antimicrobial humoral response (sensu Vertebrata)                           

GO:0045927:positive regulation of growth                                              

GO:0008015:circulation                                                                

GO:0008630:DNA damage response, signal transduction resulting in induction of apoptosis

GO:0030330:DNA damage response, signal transduction by p53 class mediator             

GO:0045892:negative regulation of transcription, DNA-dependent                        

GO:0000122:negative regulation of transcription from RNA polymerase II promoter       

GO:0000080:G1 phase of mitotic cell cycle                                             

GO:0030308:negative regulation of cell growth                                         

GO:0045792:negative regulation of cell size                                            

GO:0045926:negative regulation of growth                                              

GO:0000082:G1/S transition of mitotic cell cycle                                      

GO:0007265:Ras protein signal transduction                                            

GO:0016044:membrane organization and biogenesis                                       

GO:0016337:cell-cell adhesion                                                         

GO:0050678:regulation of epithelial cell proliferation                                

GO:0030334:regulation of cell migration                                               

GO:0030520:estrogen receptor signaling pathway                                        

GO:0007507:heart development                                                          

GO:0007160:cell-matrix adhesion                                                       

GO:0007417:central nervous system development                                         

GO:0051301:cell division                                                              

GO:0001525:angiogenesis                                                               

GO:0008544:epidermis development                                                      

GO:0006096:glycolysis                                                                 

GO:0008016:regulation of heart contraction                                            

GO:0007605:sensory perception of sound                                                 

GO:0050954:sensory perception of mechanical stimulus                                  

GO:0045765:regulation of angiogenesis                                                 

GO:0006817:phosphate transport                                                         

GO:0015698:inorganic anion transport                                                  

GO:0007156:homophilic cell adhesion                                                   

        GO:0007254:JNK cascade           
 
Back to Top

The prostate-cancer-related KEGG pathways

Similar to the above section, 13 prostate-cancer-related pathways are identified.

         path:hsa01510 Neurodegenerative Disorders
         path:hsa05030 Amyotrophic lateral sclerosis (ALS)
         path:hsa04210 Apoptosis
         path:hsa04010 MAPK signaling pathway
         path:hsa04110 Cell cycle
         path:hsa04510 Focal adhesion
         path:hsa04320 Dorso-ventral axis formation
         path:hsa04810 Regulation of actin cytoskeleton
         path:hsa04310 Wnt signaling pathway
         path:hsa04530 Tight junction
         path:hsa04620 Toll-like receptor signaling pathway

   path:hsa04920 Adipocytokine signaling pathway;

   path:hsa04350 TGF-beta signaling pathway.

 

Back to Top


Revised: March 29, 2007 .