Deep-learning approach rapidly predicts where metals bind within proteins
In the living world, roughly a third of all proteins we know rely on metals to function. Zinc helps enzymes break down molecules, iron helps carry oxygen in the blood, calcium helps relay signals in cells and potassium flows through channels that help keep our hearts beating. But despite the important role they play, scientists have long struggled to pinpoint exactly where metal ions bind in a…
Recent research from Hokkaido University has introduced a novel deep-learning technique called PRIME (Probe-based Identification of Metal-binding sites) that swiftly and accurately predicts where metal ions bind within proteins. This discovery, published in Nature Communications, addresses a longstanding challenge in biology - locating the minuscule metal-binding sites in large proteins akin to finding a needle in a haystack.
PRIME's algorithm operates in two key stages. Initially, a language model scrutinizes the protein's amino acid sequence to gauge the likelihood of each segment interacting with a metal. Subsequently, it employs virtual probes at these promising locations and assesses the immediate three-dimensional environment using an additional model to confirm the likelihood of metal binding and its precise positioning.
Evolution has preserved crucial metal-binding areas within proteins over millions of years, and PRIME leverages this preserved information embedded in current protein sequences. By merging this sequence data with extensive protein structure datasets, PRIME achieves its predictions. When tested across 14 different metal ions, including essential ones like zinc, copper, iron, sodium, and calcium, PRIME not only outperformed existing tools for transition metals but also excelled in predicting binding sites for more loosely interacting metals - areas previously difficult to predict due to their less tight interactions.
Remarkably, PRIME completes its predictions in just 11 seconds, a significant speed increase of about 10 times over current methods. The potential of PRIME extends to analyzing entire organisms or extensive protein databases, unveiling an almost hidden world of metalloproteins. A screening of 1,000 randomly selected protein families revealed that approximately 14% were confidently predicted to bind metals, with nearly 8% of these lacking annotations in existing databases.
The implications of this technology are significant, particularly for health and disease. Metal imbalances can lead to detrimental effects, such as zinc deficiency weakening immunity and iron deficiency causing anemia. PRIME's ability to predict metal-binding sites can aid in the development of drugs targeting metal-dependent proteins and potentially lead to the creation of new enzymes for industrial applications.
Looking forward, the research team aims to broaden PRIME's scope to encompass rarer and less-studied metals, ultimately conducting large-scale metalloproteome analyses to discover new and previously unknown metalloproteins.
Written by urgent.news from Phys.org's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
