A spectral framework for measuring diversity in multiple sequence alignments.

Leff, a spectral measure of effective diversity, quantifies the independent alignment positions in MSAs, revealing nearly half the diversity is constrained by functional and evolutionary factors.

• Why it matters: Understanding the true informational content in MSAs is crucial for improving machine learning models for proteins and RNAs, yet current measures lack interpretability and predictive power.
• What they did: The authors developed Leff, an interpretable spectral metric, and applied it to RNA and protein MSAs, experimental libraries, and computational datasets to assess diversity, constraints, and modeling challenges.
• The result: Leff accurately predicts protein structure modeling success, measures library novelty, and guides design efforts, establishing it as a practical tool for estimating effective information and addressing modeling limitations.