Improving the potency prediction for chemically modified siRNAs through insights from molecular modeling of individual sequence positions.

Chemical modifications are applied to small interfering RNAs (siRNAs) to improve their metabolic stability, specificity, and duration of pharmacodynamic effects. Despite tremendous progress made, identifying chemically modified siRNAs with drug-like properties requires empirical screening due to an intricate interdependence of siRNA sequence and chemistry, i.e., the nature and position of chemical modifications within the siRNA duplex. To improve our ability to design fully modified, potent siRNAs, we combined experimental measurements of thermodynamic stability and biological activity with extensive molecular modeling of the structural, dynamic, and energetic properties of parent (unmodified) siRNA duplex sequences compared with their chemically modified variants. A pattern of modifications at specific positions were identified, where the combination of sequence and chemical modifications play an outsized role in the observed biological activity. Molecular modeling revealed low stabilization energies and increased sugar stereochemical flexibility for 2'-F modified position g2 and less so for g6 in the guide strand seed region. Machine learning confirmed that these properties correlate with higher observed biological activity. These results provide molecular-level insights into the effects of chemical modifications on the intrinsic activity of siRNAs, which can be used in the rational design of chemically modified siRNAs with uncompromised potency.