Linear modeling using substructure counts

random experiments were performed, this time using a pool of 20 random pseudodescriptors, obtaining best selections of n = 1. 5 pseudodescriptors. The results are as follows  

For each of our best models, the difference between and mhrR is between 8.1 and 12.3 standard deviations, which means that the original models fit the data far 

For identical , values are higher for the SC-based models than for the Tl-based 

3 Linear modeling using both topological indices and substructure counts 

Finally, we will use the 18 topological indices and the 20 substructure counts to calculate the best linear models. Initially, we calculate the correlation matrix. One pair of completely correlated descriptors is found For all M in our real library (and for all decanes) the following is true SCi(M) = - 9. Therefore we neglect SCj. 

---