Structome-DeepScore: a distance metric derived from latent space representations

Ashar J. Malik and David B. Ascher

The Structome suite has always been driven by the power of metrics. Early versions, Structome-Q and Structome-TM, built phylogenetic trees by relying on established metrics like the Q-score and TM-score to quantify structural similarity. The next evolution, Structome-DeepRoots, enhanced this by deriving a novel metric from the abstract, high-dimensional spaces generated by deep learning embeddings. This demonstrated that a rich phylogenetic signal could be uncovered from these powerful new representations.

This resource, Structome-DeepScore now allows users to just compare two protein structures.