DeepTMHMM2 enables accurate prediction of transmembrane protein topology and subcellular location

Inside every living cell, a thin, oily barrier separates the inner world from the outside. This barrier, known as a membrane, is not just a wall but a dynamic gatekeeper that controls what enters and leaves. To function, cells rely on special proteins that sit within this membrane, acting as channels, sensors, or anchors. Some of these proteins stretch all the way through the barrier, like a tunnel boring through a mountain, while others sit on the surface or dip in only partially. Scientists have long been able to map the proteins that cross all the way through, identifying the segments that pierce the membrane and determining which way they face. However, a significant portion of these molecular structures remained a mystery: the parts that hover near the surface or dip in without ever fully crossing to the other side. Understanding the full shape and location of these proteins is crucial because their structure dictates how they work, and without a complete map, our understanding of how cells operate remains fragmented.

Researchers have now developed a new tool called DeepTMHMM2 to solve this missing piece of the puzzle. While previous methods could accurately predict the segments of a protein that span the entire membrane, they failed to identify the more subtle, partial insertions. These partial segments include re-entrant regions, which dip into the membrane and come back out without crossing, and interfacial helices, which sit along the boundary of the membrane without penetrating it fully. The new system is the first to successfully predict these non-crossing segments alongside the traditional ones. Furthermore, it goes a step further by determining exactly which of the seventeen different biological membranes in a cell a protein belongs to, rather than just guessing that it is in "a" membrane. By training on vast amounts of data, the tool learned to recognize the complex patterns that define these elusive structures, achieving high accuracy on the standard crossing segments while simultaneously mastering these previously overlooked features.

When the researchers applied this new tool to a massive database of known proteins, the results revealed a surprising truth about the architecture of life. They found that these non-crossing segments are not rare anomalies but a common feature of the transmembrane proteome. In fact, the analysis showed that interfacial helices are present in nearly a quarter of all the proteins that cross the membrane in a spiral shape. This discovery changes the picture of how these proteins are built, showing that they are far more complex and varied than previously thought. The tool successfully learned to predict these additional elements, proving that the boundaries of membrane proteins are not just simple lines of entry and exit, but include a rich landscape of partial interactions that are essential for biological function.