A data set comprising 27 $myo$-inositol derivatives based on tetrakisphosphates and bispyrophosphates were used in the development of quantitative structure–activity relationship model for investigating its allosteric effector property against human haemoglobin (Hb). Three-dimensional structures of the investigated compounds were subjected to geometry optimisations at the density functional theory level. Physicochemical features of low-energy conformers were represented by quantum chemical and molecular descriptors. Feature selection by means of unsupervised forward selection and stepwise linear regression resulted in a set of four important descriptors. Multivariate analysis was performed using multiple linear regression (MLR), artificial neural network (ANN) and support vector machine (SVM). Robustness of the predictive performance of all methods was deduced from internal and external validation, which afforded $Q_{CV}^2$ values of 0.6306, 0.7484 and 0.8722 using MLR, ANN and SVM, respectively, for the former and $Q{Ext}^2$ values of 0.8332, 0.8847 and 0.9694, respectively, for the latter. The predictive model is anticipated to be useful for further guiding the rational design of robust allosteric effectors of human Hb.