Capturing Protein Domain Structure and Function Using Self-Supervision on Domain Architectures

Predicting biological properties of unseen proteins is shown to be improved by the use protein sequence embeddings. However, these embeddings have caveat that metadata do not exist for each amino acid, in order measure quality unique learned embedding vector separately. Therefore, current cannot intrinsically evaluated on degree their captured information a quantitative manner. We address this drawback our approach, dom2vec, learning representation domains and acid base, as domain To perform reliable intrinsic evaluation terms biology knowledge, we selected related most distinctive characteristics domain, which are its structure, enzymatic, molecular function. Notably, dom2vec obtains an adequate level performance assessment—therefore, can draw analogy between local linguistic features natural languages structure function architectures. Moreover, demonstrate applicability prediction tasks, comparing it with state-of-the-art three downstream tasks. show outperforms toxin enzymatic comparable cellular location prediction.