#40 · Structural bio & protein design

Predicted structures and ensembles used as optimization criteria in biomolecular design

The shared thread is that predicted structure or ensemble behavior is being used to steer or filter biomolecular design rather than serving only as a final readout. In the enzyme study, a PLM-derived energy landscape is sampled with BAGEL to generate miniaturized enzyme sequences, which are then screened by structure prediction and molecular dynamics. FoldARE similarly converts in silico RNA ensembles into pseudo-SHAPE constraints that guide secondary structure prediction. The confidence-integrity screen addresses the reliability of this practice: across 27,098 AlphaFold2 predictions for 18 fold-switching proteins, 33.6% high-confidence false predictions are reported. This item is an evaluation rather than a design method, so the thread applies to it only as a caution. What is new is the coupling of structure-derived quantities to optimization, which makes pLDDT-type confidence and ensemble outputs load-bearing rather than descriptive.