Further Improvements of Drug-Likeness Prediction

Both methods presented are based on statistical analyses of a large set of several thousands of proven dmg molecules. Therefore, both methods are only able to identify probable drugs by identifying their similarity to drug molecules already in existence. Unfortunately, most of these were developed many years ago, which is why current trends in drug development are missed. Shifts in the composition of drugs, such as use of other structural patterns or functional groups, are not con- 

Some characteristics, such as the increase of the mean molecular weight of new drug candidates, can be included easily by relaxing the ranges for these properties. (However, in so doing, the computational chemist abandons the soHd statistical base ) 

---