Omics-scale decoding of protein N-glycosylation using protein language model embeddings
Protein language models (pLMs) learn rich sequence semantics from vast volume of protein sequences, yet the information encoded in their high-dimensional embeddings remains largely a black box, posing an interpretability challenge. Here we show that semantic features learned by pLMs can be decoded to reveal biologically meaningful signals using coevolutionary analysis without downstream…
Protein language models (pLMs) utilize vast protein sequence data to learn complex semantic information, yet the meaning behind their high-dimensional embeddings remains unclear, creating an interpretability issue. This study demonstrates that semantic features gleaned from pLMs can be decoded through coevolutionary analysis without the need for supervised training.
This discovery led to the creation of MaskNGly, a zero-shot framework capable of identifying protein N-glycosylation sites directly from pLM-embeddings. The performance of MaskNGly is comparable to the best supervised predictors available.
The framework is both computationally efficient and intrinsically transferable, allowing for an omics-scale analysis of protein glycosylation landscape. In their investigation, they discovered that canonical N-X-S/T motifs become more prone to glycosylation as organisms evolve in complexity. Moreover, glycosylation tends to occur predominantly at structured, solvent-exposed regions of proteins. Furthermore, alterations to high-scoring motifs in proteins are more frequently associated with diseases.
The framework also uncovers signatures of glycosylation-mediated protein allostery. Overall, these findings provide an interpretable and scalable approach to studying glycobiology using pLM-embeddings.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
