You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: README.md
+27-15Lines changed: 27 additions & 15 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -23,19 +23,41 @@ A detailed User Manual can be found [here](https://github.com/DataFusion4NetBio/
23
23
Supervised Complex Detection (SCODE) is available through the Cytoscape App Store [here](http://apps.cytoscape.org/apps/scode).
24
24
To install, either
25
25
1. Open Cytoscape on your machine, and navigate to Apps > App Manager from the menu bar. Search for 'SCODE' and click Install.
26
-
2. Download the .jar file from the [Cytoscape App Store website](http://apps.cytoscape.org/apps/scode). Then move the jar file into your ~/Cytoscape/3/apps/installed/ directory.
26
+
2. Download the .jar file from the [Cytoscape App Store website](http://apps.cytoscape.org/apps/scode). Then move the jar file into your ~/Cytoscape/installed/ directory.
27
27
28
28
####Basics
29
29
To use this app:
30
30
1. Load the graph on which you would like to search for complexes.
31
31
2. Go to Apps > SCODE > Open SCODE. The application will open in the left panel.
32
-
3. Customize the training and search parameters.
33
-
4. Load training data (see below for formatting requirements).
32
+
3. Customize the search and scoring parameters.
33
+
3a. If scoring using only edge weight information, select the "no learning" option, or
34
+
3b. If scoring using a Bayesian network, load positive training file with which to train the Bayesian template (see below for formatting requirements), or select the trained Bayesian network
34
35
5. Click Analyze.
35
-
6. Results are stored as subnetworks of the target network in your session file, complete with a likelihood score. The app will also generate a new 'Model' network that represents a trained model, which you can re-use with future searches.
36
+
6. Results are stored as subnetworks of the target network in your session file, complete with a likelihood score. If you select the learning scoring option, the app will also generate a new 'Model' network that represents a trained model, which you can re-use with future searches.
37
+
7. Optional: Evaluate your results using a testing file of all known protein complexes in the network
38
+
39
+
36
40
***
37
41
38
-
#####Training the Model
42
+
#####Searching a PPI Graph
43
+
44
+
Currently, SCODE supports an [iterative simulated annealing search](http://en.wikipedia.org/wiki/Simulated_annealing) for finding candidate complexes within a dataset. This search comes in three flavors:
45
+
* ISA: This is the fastest option and will perform the worst. Each round, a candidate is expanded (or not) using a single, random neighboring node.
46
+
* M-ISA: This a slower option that will perform better than ISA. Each round, a candidate is expanded by testing the M highest degree neighboring nodes. The best of these M nodes is used for expansion.
47
+
* Greedy-ISA: This option, the slowest, tests all the neighboring nodes for expansion and selects the best one. This will result in more, larger, higher-scoring candidate complexes.
48
+
49
+
SCODE allows you to specify several additional search parameters to define the scope of the search (temperature, scaling ratio, etc). For an explanation of each of these parameters, see the more detailed [User Manual](Demo/SCODEUserManual.pdf)
50
+
51
+
#####Scoring Candidate Complexes
52
+
There are three options for scoring candidate complexes:
53
+
1. Score using only edge weights (no learning)
54
+
2. Train a Bayesian template with positive complex exemplars
55
+
3. Score using a trained Bayesian model
56
+
57
+
Option 1 does not require any additional input information; the average edge weight among the members of a candidate complex will be used to calculate its score.
58
+
Options 2 and 3 are described in more detail below.
59
+
60
+
#####Training a Bayesian Template
39
61
The type of model used by SCODE is a [Bayesian network](http://en.wikipedia.org/wiki/Bayesian_network). Bayesian networks are probabilistic graphical models that make it easy to define relationships between supposed features of complexes. Each node in a Bayesian network represents a feature (e.g. Number of nodes in a complex). Each edge between nodes represents a dependency between features or conditioning of one feature by another (e.g. the complex's density given the number of nodes in the complex). The values of each feature are discretized before training or scoring a candidate complex.
40
62
41
63
SCODE provides several options for creating or loading a model. You may:
@@ -53,16 +75,6 @@ Each row represents a known complex, with tab-separated columns as follows:
53
75
54
76
***
55
77
56
-
#####Searching a PPI Graph
57
-
58
-
Currently, SCODE supports an [iterative simulated annealing search](http://en.wikipedia.org/wiki/Simulated_annealing) for finding candidate complexes within a dataset. This search comes in three flavors:
59
-
* ISA: This is the fastest option and will perform the worst. Each round, a candidate is expanded (or not) using a single, random neighboring node.
60
-
* M-ISA: This a slower option that will perform better than ISA. Each round, a candidate is expanded by testing the M highest degree neighboring nodes. The best of these M nodes is used for expansion.
61
-
* Greedy-ISA: This option, the slowest, tests all the neighboring nodes for expansion and selects the best one. This will result in more, larger, higher-scoring candidate complexes.
62
-
63
-
SCODE allows you to specify several additional search parameters to define the scope of the search (temperature, scaling ratio, etc). For an explanation of each of these parameters, see the more detailed [User Manual](Demo/SCODEUserManual.pdf)
0 commit comments