Kohonen SOM (.som)

Kohonen SOM (.som)

The SOM file records the output of the Kohonen mapping. The Kohonen map is a square table of cells or clusters. The data points from the input file are distributed to the map cells, such that similar data points will fall in the same cell or in nearby cells. As a result, each cell in the map contains zero or more compounds.

The SOM file contains:

A Header

Program Identifier <size of the Kohonen map>

Cell information

(row, col)= [<index> <index>…] : <average dependent variable values of this cell>

Compounds

Their cell assignment and whether they were allocated to the train, verify or test set.

The seven columns in this section include following column headers

1. NUMBER: the index assigned to the compound

2. NAME: the name of the compound

3. SET: the set that the compound was placed in (train, verify or test)

4. CELL: the cell number to which the compound was mapped (zero-based)

5. CLUSTER SIZE: the total number of compounds that were mapped to the same cell.

6. ROW: the row number of the cell (zero-based).

7. COLUMN: The column number of the cell (zero-based)

The cell number is equal to (row * number of cells per side + column).

A sample SOM file looks like this (modified for clarity):

# ADMET Modeler(TM) 0.0.0020 Kohonen map:     8

(0,0)= 7    6    5    :    0.30860095976375

(0,1)= 31    33    :    0.655592469545958

(0,2)= :    0

(0,3)= 59    :    0.946843853820598

(0,4)= :    0

…

(6,6)= :    0

(6,7)= 2    :    0.08859357696567

…

(7,6)= :    0

(7,7)= :    0

NUMBER    NAME    SET    CELL    CLUSTER SIZE    ROW    COLUMN

00000    Data1    Test    14    6    1    6

00001    Data2    Test    14    6    1    6

00002    Data3    Train    55    1    6    7

00003    Data4    Train    44    2    5    4

00004    Data5    Train    32    1    4    0