ANNE Performance (.ann)

The file consists of a header block followed by a block of data describing the final set of networks included in ANNE performed individually. The networks are sorted in order of increasing merit function and performance statistics that compare the predicted to the observed outputs for training, verify and test sets are provided for each.

Header Block

Program Identifier <Nets per ANNE>

ANN Block for each network

Merit: <merit after optimization> <training RMSE > <verify RMSE > <test RMSE >

QSQD: <training set Q2 value> <verify set Q2 value > <test set Q2 value >

SRCC: <training set SRCC value > <verify set SRCC value > <test set SRCC value >

FALSERATE: <training set FalseRate> <verify set FalseRate> <test set FalseRate>

Training set size= <number of data points in the training set>

< index1> <index2> ... <indexn>

Verify set size= <number of data points in the verify set>

< index1> <index2> ... <indexn>

where:

RMSE is the Root-Mean-Square Error

QSQD is the Coefficient of Determination (Q²)

SRCC is Spearman's Rank Correlation Coefficient between predictions and observations

For classification models, YOUDEN and MCC replace QSQD and SRCC.

YOUDEN is average of the sum of sensitivity and specificity normalized by class size.

MCC is the generalized Matthews Correlation Coefficient

The index of the compound is for the original data file.

Part of a typical ANN file is shown below. Ellipses (...) indicate where lines have been omitted.

ADMET Modeler(TM) 4.1.0055 ANN file:

c:\...\Example3 5/21/2010 4:34:08 PM99

36 3

Merit: 0.157803 0.315606 0.311546 0.319881

QSQD: 0.71056 0.693718 0.381251

SRCC: 0.776802 0.81742 0.598063

FALSERATE: -977.000 -977.000 -977.000

none

Training set size= 26

2 3 8 9 18 19 20 23 24 25 …

Verify set size= 19

4 6 7 10 12 13 15 27 28 37 …

36 3

Merit: 0.172813 0.345627 0.339334 0.253916

QSQD: 0.647781 0.648087 0.610133

SRCC: 0.777471 0.746991 0.658006

FALSERATE: -977.000 -977.000 -977.000

none

...