Variable D i is the new SD value after excluding the outliers in scene s i, and α is a constant predictive coefficient.
The best CoMFA model yielded a predictive correlation coefficient r2pred value of 0.809.
The predictive power of the developed model was obtained using a test set of 7 compounds, giving a predictive correlation coefficient r2pred of 0.767.
For the test set, it gave a predictive correlation coefficient (R) of 0.955, root mean square error (RMSE) of 1.041 and absolute average relative deviation (AARD, %) of 4.663, respectively.
Among all the built models the most predictable one is constructed based on the simplest combination of heuristic variable selection and MLR modeling, with its predictive coefficient of determination (r2pred) and root-mean-square error of prediction (RMSP) on external independent test set of 0.90 and 0.39, respectively.
When the same procedure was applied to data from the Class II pharmacophore model, a combination of Tyr, Met, Val, Val, Thr, Leu, Phe and Ile amino acid residues yielded the best statistical correlation coefficients r2 = 0.92 and cross-validated predictive coefficient q2 = 0.43.
When the ICU admission diagnosis was unavailable or not coded by the automatic tool, the corresponding predictive coefficient was replaced by the 'generic' adjusted diagnosis coefficient of −0.42772.
In addition, the predictive coefficient of age in these formulas use constant values throughout the life span in the same ethnic group, 3, 9, 10, 19 which may have distorted and produced prediction errors from different spectrums in the age groups due to biological distinctions in metabolic synthesis in older people.
These include traditional features such as linear predictive coefficients, linear prediction cepstral coefficients, mel-frequency cepstral coefficients (MFCC [2], or unconventional ones like perceptual linear predictive coefficients, log frequency power coefficients[3], gammatone frequency cepstral coefficients[4], or compact multiclass support vector machines[5].
Annual mean mortality rates have been correlated with ambient aerosol concentration indices, with SO4 2-), FP, IP, and TSP having a descending order as predictive coefficients.
