CHEMSPREAD

CHEMSPREAD is a spread-sheet-type statistical analysis program designed for correlation of numerical data associated with chemical structures.

When applied to the results of calculations using MOPAC, special output files (type *.pro) produced by the latter program are passed through the auxiliary program SPREAD to produce an input file for CHEMSPREAD (file type *.prs).

• DISPLAY FORMAT. The program as supplied is capable of handing up to 200 properties (columns) associated with up to 200 structures (rows), but can easily be re-dimensioned to accomodate larger or smaller problems. At any one time up to seven data columns (plus a column of compound identifiers) and up to 41 rows (plus column headings) may be displayed. Scrolling up/down and left/right reveals all of the data entries .
  
• PROGRAM CONTROL.
  • This is achieved using a permanent menu column together with pop-up menus.
• BASIC FACILITIES.
  • Input and output of data sets in three formats.
    • EXCEL format
    • ARTHUR format
    • SPREAD format
  • Facilities to add single rows in SPREAD format
  • Provision to allocate a column as a "non-numeric" classifier (category)
  • Move row
  • Move column
  • Delete columns
  • Delete rows
  • Delete all marked columns and/or rows
  • Sort data entries (on basis of the values in a column)
  • Display of data
    • Original data
    • Normalised data
    • Mean-centred data
    • Auto-scaled data
  • Display column statistics (maximum, mimimum, range, mean, standard deviation, second moment, third moment, fourth moment, skewness, kurtosis).
  • Display column correlations
  • Display column covariance
  • Display row-row cartesian distances
  • 2D plots of pairs of columns of data
  • 3D plots of triplets of columns of data
    
• DATA PRUNING
  • The data may be analysed to achieve a reduction in the number of columns by several processes:-
    • Eliminate columns with high Kurtosis
    • Eliminate columns with zero range
    • Eliminate columns very highly correlated to other columns
    • Progressive elimination of columns which are highly correlated to other columns
  • Columns are initially "marked for deletion" and appear in the display "crossed out"
  • Subsequently marked columns are deleted
    
• PRINCIPAL COMPONENT ANALYSIS
  • Standard principal component analysis techniques can be used
    • Choice of correlation matrix or covariance matrix
    • Display communalities
    • Varimax rotation
    • Orthogonalisation of rotated matrix
    • 2D plot of pairs of residual eigenvectors
    • 3D plot of triplets of residual eigenvectors
  • Use of Non-linear Partial Least Squares method (NIPALS)
    • Choice of autoscaled or mean-centred data matrix
      
• COMPOUND CLUSTERING
  • Hierarchical clustering - choice of:-
    • Cartesian distance matrix,
    • Tanimoto coefficient matrix
    • Cosine coefficient matrix
    • Dendrogram construction
  • Jarvis-Patrick clustering
  • Reciprocal-nearest-neighbour clustering
  • K-means clustering
    
• NON-LINEAR MAPPING
  • Reduce dimensionality of data to 2- or 3-dimensions
• 
  
• AUXILIARY PROGRAM
  • The program SPREAD is supplied to facilitate analysis of results produced by MOPAC

Back to INTERPROBE Home Page

Revised 18th June 2007