Gast ::
| Hauptseite > Publikationsdatenbank > RNA fitness prediction with sparse physics-based models — a way to explore the sequence space |
| Hauptseite > Publikationsdatenbank > RNA fitness prediction with sparse physics-based models — a way to explore the sequence space |
| Poster (After Call) | FZJ-2025-05455 |
; ; ;
2025
Abstract: The field of medicine uses macromolecules as a means of therapeutic intervention. Consequently, the functional attributes of these novel molecules are assuming greater significance. To complement the extensive wet-lab experiments, we have devised a series of statistical physics based models that are capable of predicting the fitness of RNA molecules based on one- and two-point mutation scans.The experimental data were employed as training data to fit models of increasing complexity, commencing with an additive model and concluding with a model that accounts for global and local epistasis. The trained models were validated using fitness data from scans with more than two point mutations of the wild-type. In contrast to conventional AI algorithms, the parameters of our models were designed to facilitate direct interpretation.In examining more distant sequences, we can distinguish the corresponding RNA family from random sequences with a high degree of accuracy. Moreover, the models facilitate direct interpretations of evolutionary processes and the significance of epistatic terms. Our model can be used to create a fitness landscape far beyond the experimentally sampled sequence space, thus identifying promising RNA molecules. Furthermore, the extension to the entire sequence space can be used as a blueprint for other molecules, providing a novel avenue for questions in biomolecular design.
|
The record appears in these collections: |