Transport proteins play a crucial role in constantly moving substances in and out of biological cells. Identifying which specific substances a transport protein can handle has historically been a challenge. However, bioinformaticians have created a predictive model named SPOT that leverages artificial intelligence (AI) to address this issue with high precision.
Transport proteins are essential for facilitating the continuous movement of substances in and out of biological cells. Determining which specific substances a transport protein can handle proves to be challenging. Researchers from Heinrich Heine University Düsseldorf (HHU) have introduced a model called SPOT, which utilizes artificial intelligence (AI) to predict transport capabilities with remarkable accuracy. Their findings have been shared in the scientific publication PLOS Biology.
To ensure cell survival and functionality, substrates within biological cells must be consistently transported across the cell membrane. However, not every substance that circulates in the body can enter the cells freely. Furthermore, some transport mechanisms need to be regulated, activating only during specific conditions or times to trigger certain cellular functions.
Transport proteins, often referred to as transporters, serve as the gatekeepers for these specialized transport channels. They are embedded in cell membranes and are composed of numerous amino acids that together create a complex three-dimensional structure.
Each transporter is customized for a distinct molecule, known as the substrate, or a small collection of substrates. Yet, the question remains: which substrates are suited for each transporter? Researchers are continually investigating the correct transporter-substrate matches.
According to Professor Dr. Martin Lercher from the Computational Cell Biology research team, “Testing which substrates correspond to which transporters can be a challenging experimental task. Even capturing the three-dimensional structure of a transporter can be complex since these proteins tend to lose stability once separated from the cell membrane.”
“We opted for an alternative approach based on AI,” explains Dr. Alexander Kroll, the lead author and postdoctoral researcher in Lercher’s lab. “Our method, called SPOT, has utilized over 8,500 experimentally validated transporter-substrate pairs as training data for a deep learning model.”
The Düsseldorf bioinformaticians convert protein sequences and substrate molecules into numerical vectors, making them compatible with AI models. Once the model is trained, researchers can input vectors for new transporters and potential substrates into the AI system, which then predicts the likelihood of a substrate matching a transporter.
“Kroll states: “We validated our model using an independent dataset with known transporter-substrate pairs. SPOT demonstrates an impressive accuracy of over 92% in predicting whether a given molecule is a substrate for a specific transporter.”
SPOT efficiently proposes promising substrate options, significantly narrowing the focus for lab researchers. “This effectively accelerates their ability to discover the correct substrate for a transporter,” explains Professor Lercher, emphasizing the connection between bioinformatics predictions and experimental validation.
Kroll adds, “This technology applies to any transport protein, not just limited types, unlike previous models.”
There are extensive potential applications for this model. Lercher notes, “In biotechnology, we can alter metabolic pathways to produce specific items like biofuels, or we can design drugs to target specific transporters for more effective cell delivery.”
