Empirical methods for computing molecular partition coefficients: II. Inclusion of conformational flexibility within fragment–based approaches