ROOT

1

• Use Analyze root 34 for easy way
• (the file meta let you override defaults)
• Use meta root for full mode
• - e.g meta root
• - use MetaUI for input file

ROOT ROOT.PAT ROOT.TES (ROOT.WGT) (ROOT.FWT) (ROOT
.DBD)
MetaNeural
ROOT.XXX ROOT.TTT ROOT.TRN (ROOT.DBD) ROOT.WGT ROO
T.FWT
2
ANALYZE MetaNeural Alternative Code Either
run meta root analyze root.pat
34 (single training and testing)
analyze root.pat 3434 (LOO)
analyze root.txt 34 (bootstrap mode) Results
for analyze are in resultss.xxx and
resultss.ttt Results from MetaNeural are in
root.xxx and root.ttt MetaNeural input file is
generated automatically in analyze The file
name meta overrides the default input file for
analyze
3
MetaNeural Input File for the ROOT
4 gt 4 layers 2 gt 2 inputs 16
gt hidden neurons in layer 1 4 gt
hidden neurons in layer 2 1 gt
outputs 300 gt epoch length (hintalways use 1,
for the entire batch) 0.01 gt learning parameters
by weight layer (hint 1/ patterns or 1/
epochs) 0.01 0.01 0.5 gt momentum parameters
by weight layer (hint use 0.5) 0.5 0.5 10000000
gt some very large number of training epochs
200 gt error display refresh rate 1
gtsigmoid transfer function 1
gt Temperature of sigmoid check.pat
gt name of file with training patterns (test
patterns in root.tes) 0 gt not
used (legacy entry) 100 gt not used
(legacy entry) 0.02000 gt exit training if
error lt 0.02 0 gt initial weights
from a flat random distribution 0.2
gt initial random weights all fall between 2 and
2
4
EXAMPLE DATA SETS

• IRIS data
• Checkerboard data
• Svante wolds QSAR data
• Cherkasskys nonlinear function
• Albumin QSAR data

5
(No Transcript)
6
FILES RELATED TO CHECKERBOARD EXAMPLE
CHECK_NET.BAT
CHECK_DATA.BAT
CHECK_TEST.BAT
CHECK.PAT
7
MetaNeural INPUT FILE FOR CHECKERBOARD DATA
8
(No Transcript)
9
(No Transcript)
10
(No Transcript)
11
QSAR DATA SET EXAMPLE 19 Amino Acids
From Svante Wold, Michael Sjölström, Lennart
Erikson, "PLS-regression a basic tool of
chemometrics," Chemometrics and Intelligent
Laboratory Systems, Vol 58, pp. 109-130 (2001)
RENSSELAER
12
(No Transcript)
13
PLS 1 latent variable
14
PLS 1 latent variable No aromatic AAs
15
1 latent variable Gaussian Kernel PLS (sigma
1.3) With aromatic AAs
16
Chemoinformatic Models to Predict Binding
Affinities to Human Serum Albumin G. Colmenarejo
et. al., J. Med. Chem 2001, 44, pp. 4370-4378
95 Molecules Widely different compounds 250-1500
Descriptors
17

• Binding affinities to human serum
• albumin (HSA) log Khsa
• Gonzalo Colmenarejo, GalaxoSmithKline
• J. Med. Chem. 2001, 44, 4370-4378
• 95 molecules, 250-1500 descriptors
• Widely different compounts

18
Electron Density-Derived TAE-wavelet Descriptors

• 1 ) Surface properties are encoded on 0.002
  e/au3 surface
• Breneman, C.M. and Rhem, M., J. Comp. Chem.,
  1997,18(2), p. 182-197
• 2 ) Histograms or wavelet encoded of surface
  properties give TAE property descriptors

19
PEST-Shape Descriptors Surface Property-Encoded
Ray Tracing

• TAE Internal Ray Reflection - low resolution scan

Isosurface (portion removed) with 750 segments
RENSSELAER
20
Shape-Aware Molecular Descriptors
from Property/Segment-Length Distributions

• Segment length and point-of-incidence value form
  2D-histogram
• Each bin of 2D-histogram becomes a hybrid
  descriptor
• 36 descriptors per hybrid length-property

PIP vs Segment Length
RENSSELAER
21
training
22
testing
23
(No Transcript)
24
(No Transcript)
25
(No Transcript)
26
CHERKASSKYS NONLINEAR BENCHMARK DATA
Generate 500 datapoints (400 training 100
testing) for
Cherkas.bat
27
(No Transcript)
28
(No Transcript)
29
(No Transcript)
30
Ysinx/x
Generate 500 datapoints (100 training 500
testing) for
31
Comparison Kernel-PLS with PLS 4 latent
variables sigma 0.08
PLS
Kernel-PLS
32
(No Transcript)