Learning the structural diversity in random protein sequence space

Random protein sequence libraries contain multiple physical phenotypes beyond soluble versus aggregated, including sequences yielding substantial FRET in both cytosol and visible assemblies.

• Why it matters: Life’s sampled protein sequences are a tiny fraction, leaving unknown which physical properties unevolved polypeptides already access versus those requiring evolutionary refinement.
• What they did: Using high-throughput FRET-based selection on a library of millions of random protein sequences, plus fluorescence imaging, biophysical characterization, and sequence-based modelling.
• The result: Phenotypes are predictable from sequence, and repeated selection for stronger FRET reduces population solubility, showing proximity-based selection alone may not favor biology-like protein states.