Informatics calibration of a molecular descriptors database to predict solid dispersion potential of small molecule organic solids

Abstract

The use of a novel, in silico method for making an intelligent polymer selection to physically stabilize small molecule organic (SMO) solid compounds formulated as amorphous molecular solid dispersions is reported. 12 compounds (75%, w/w) were individually co-solidified with polyvinyl pyrrolidone:vinyl acetate (PVPva) copolymer by melt-quenching. Co-solidified products were analyzed intact using differential scanning calorimetry (DSC) and the pair distribution function (PDF) transform of powder X-ray diffraction (PXRD) data to assess miscibility. Molecular descriptor indices were calculated for all twelve compounds using their reported crystallographic structures. Logistic regression was used to assess correlation between molecular descriptors and amorphous molecular solid dispersion potential. The final model was challenged with three compounds. Of the 12 compounds, 6 were miscible with PVPva ( i.e. successful formation) and 6 were phase separated ( i.e. unsuccessful formation). 2 of the 6 unsuccessful compounds exhibited detectable phase-separation using the PDF method, where DSC indicated miscibility. Logistic regression identified 7 molecular descriptors correlated to solid dispersion potential ( α = 0.001). The atomic mass-weighted third-order R autocorrelation index (R3m) was the only significant descriptor to provide completely accurate predictions of dispersion potential. The three compounds used to challenge the R3m model were also successfully predicted.