Friday, March 23, 2012

COG 50000 Blast Untransformed with Chisq1

Description

DataSet: COG Size: 50000 Unique: Yes
Aligner: Blast ScoringMatrix: BLOSUM62 GapOpen: -16 GapExt: -4
DistanceType: Normalized BitScore (2*AB/AA+BB) Transformation: None
Mapping: Chisq1 DistanceCut: 0.96
Initialization: Random
Fixed: None
Varied: All
DensitySat: 0.85

Links

Images


Full Sample with Selected Clusters


Configuration

I/O
 CoordinateWriteFrequency:      0
 DistanceMatrixFile:            F:\Salsa\saliya\cog\100k\input\cog_95672_bitscore_refined_first50k_c#.bin


ManxcatCore
 AddonforQcomputation:          2
 CalcFixedCrossFixed:           True
 CGResidualLimit:               1E-05
 ChisqChangePerPoint:           0.001
 Chisqnorm:                     1
 ChisqPrintConstant:            1
 ConversionInformation:         
 ConversionOption:              
 DataPoints:                    50000
 Derivtest:                     False
 DiskDistanceOption:            2
 DistanceCut:                   0.96
 DistanceFormula:               1
 DistanceProcessingOption:      0
 DistanceWeigthsCuts:           
 Eigenvaluechange:              0.001
 Eigenvectorchange:             0.001
 ExtraOption1:                  0
 Extraprecision:                0.05
 FixedPointCriterion:           none
 FletcherRho:                   0.25
 FletcherSigma:                 0.75
 FullSecondDerivativeOption:    0
 FunctionErrorCalcMultiplier:   10
 HistogramBinCount              100
 InitializationLoops:           1
 InitializationOption:          1
 InitialSteepestDescents:       0
 LinkCut:                       5
 LocalVectorDimension:          3
 Maxit:                         80
 MinimumDistance:               -0.001
 MPIIOStrategy:                 0
 Nbadgo:                        6
 Omega:                         1.25
 OmegaOption:                   0
 PowerIterationLimit:           200
 ProcessingOption:              100
 QgoodReductionFactor:          0.5
 QHighInitialFactor:            0.01
 QLimiscalecalculationInterval: 1
 RotationOption:                0
 Selectedfixedpoints:           
 Selectedvariedpoints:          
 StoredDistanceOption:          2
 TimeCutmillisec:               -1
 TransformMethod:               0
 TransformParameter:            0.125
 UndefindDistanceValue:         -1
 VariedPointCriterion:          all
 WeightingOption:               0
 Write2Das3D:                   True


Density
 Alpha:                         2
 Pcutf:                         0.85
 SelectedClusters:              63,82,145,265,362,684,708
 XmaxBound:                     1.8
 Xres:                          50
 YmaxBound:                     1.8
 Yres:                          50

1 comment:

  1. There are two things one tries to do
    a) Transform input distance distribution so that one preserves order i.e. distance --> function(distance)such that f(d1) > f(d2) if d1 > d2. Transform Method 10 and 4D transformation aim at this

    b) weight contributions to Chisq sum of (euclidean mapped distance - input distance)^2
    Here SMACOF has weight 1
    Sammon has weight 1/distance
    Chisq1 refers to parameter Chisqnorm = 1 which is weight 1/distance^1.5

    Sammon (chisqnorm 2) and Chisq1 enhance small distances compared to traditional SMACOF

    ReplyDelete