The challenge of experimentally characterizing transporters is formidable. Membrane proteins are notoriously unstable once removed from their native lipid environment, which complicates purification, crystallization, and nearly every classical biochemical assay used to establish what a protein binds. As a result, curated databases of experimentally verified transporter–substrate relationships cover only a small fraction of known transporters, and many solute carrier proteins—the largest transporter family in humans—remain so-called orphans with no assigned substrate. Computational tools could in principle fill this gap, but existing methods have tended to focus on narrow classes of substrates or have struggled to generalize to transporters that are evolutionarily distant from those in their training data.

TraSPIN, developed by Tong Pan, Fuyi Li, Jiangning Song, and colleagues at institutions including the South Australian immunoGENomics Cancer Institute, Monash University, and the University of Melbourne, takes a fundamentally different approach. The framework is built on geometric deep learning, a branch of machine learning designed to operate on graph-structured data, and it jointly models both sides of the recognition problem: the transporter protein and the small-molecule substrate. By representing transporters and substrates in a shared learning framework, the model can learn the relational patterns that connect a protein’s sequence and structure to the chemical features of the molecules it accepts.

A central innovation is the use of protein language models as the foundation for transporter representations. These models, analogous to the large language models behind modern chatbots, are trained on enormous databases of protein sequences and learn to encode the statistical grammar of amino acid chains. In doing so, they capture subtle information about which positions in a sequence are functionally constrained, which residues are likely to form binding pockets, and how mutations are tolerated. TraSPIN leverages these learned embeddings to build rich, transferable representations of transporters, which are then combined with graph-based learning over protein–substrate interaction networks. The graph structure allows the model to propagate information across related transporters and related substrates, so that knowledge about one interaction informs predictions about others.

This architecture matters because specificity prediction is, at its core, a link-prediction problem on a sparse and biased network. The known interaction map is dense only for a handful of well-studied transporter families and substrate classes, such as sugar transporters or neurotransmitter carriers. A model that simply memorizes these dense regions performs poorly when asked about a transporter from an unexplored family. TraSPIN’s language-model-guided design is explicitly aimed at this generalization problem. In benchmark comparisons reported in the paper, the framework outperformed existing computational methods, and critically, it retained predictive accuracy on substrates it had never seen during training and on transporters whose sequences are dissimilar to anything in the training set—precisely the scenarios where earlier tools broke down.

Beyond raw prediction accuracy, the authors show that TraSPIN is interpretable in biologically meaningful ways. The model’s internal representations can be interrogated to reveal structural determinants of transporter–substrate recognition and sequence motifs associated with binding particular classes of molecules. In other words, the network does not behave as an inscrutable black box; it appears to have learned features that map onto real biochemical principles of molecular recognition. This kind of mechanistic insight is valuable in its own right, because it can guide experimentalists toward specific residues or structural regions to mutate and test, turning a purely computational prediction into a testable hypothesis.

The practical demonstrations in the study highlight two applications with immediate relevance to biology and medicine. The first is cross-species inference: using knowledge accumulated about transporters in well-characterized organisms to predict substrate specificities in species where little experimental data exists. This matters for microbiology, parasitology, and biotechnology, where transporter function in non-model organisms often underpins nutrient uptake, drug resistance, and metabolic engineering. The second application is de-orphanization—the assignment of substrates to uncharacterized solute carrier proteins. Human solute carriers regulate drug absorption, metabolism, and disposition, and orphan carriers are increasingly recognized as potential therapeutic targets and as mediators of unexplained drug–drug interactions. A reliable computational tool that can propose substrates for these orphans could dramatically accelerate functional annotation efforts that would otherwise require years of laboratory work.

The broader context is a rapidly growing intersection between artificial intelligence and structural biology. Protein language models have already transformed fields from structure prediction to protein design, and geometric deep learning has proven adept at tasks where the geometry of molecules—whether the three-dimensional fold of a protein or the two-dimensional scaffold of a small molecule—carries the essential signal. TraSPIN sits squarely at this confluence, treating transporter specificity not as an isolated classification task but as a relational inference problem across a heterogeneous graph of proteins and chemicals. The authors’ results suggest that this joint modeling strategy, rather than modeling proteins and substrates separately, is key to the framework’s generalization ability.

There are, of course, the usual caveats that accompany any machine learning advance in biology. Predictions of transporter–substrate interactions are hypotheses, not measurements, and the sparsity of ground-truth data means that performance estimates on rare substrate classes carry uncertainty. The authors acknowledge that experimental validation remains essential, and the model’s greatest value may lie in prioritizing which of thousands of untested transporter–substrate pairs should be examined first in the laboratory. Used this way—as an intelligent filter rather than an oracle—computational frameworks like TraSPIN can multiply the productivity of experimental programs focused on membrane proteins, where each assay is technically demanding and expensive.

Nevertheless, the arrival of TraSPIN marks a significant step toward solving one of the most persistent annotation problems in molecular biology. The genomes of humans and of every model organism encode hundreds of transporters whose substrates have never been determined, and each uncharacterized carrier is a potential link in a metabolic pathway, a drug target, or a determinant of drug response. By combining the sequence-reading power of protein language models with the relational reasoning of graph learning, TraSPIN offers a scalable route from sequence to function for this protein universe. If its predictions continue to hold up under experimental scrutiny, the era in which transporter specificity is inferred by computer before it is confirmed at the bench may arrive sooner than anyone expected.

Subject of Research: Computational prediction of membrane transporter substrate specificity using protein language models and geometric deep learning

Article Title: TraSPIN: inferring transporter specificity through language model-guided graph learning

Article References: TraSPIN: inferring transporter specificity through language model-guided graph learning. (n.d.). https://doi.org/10.1186/s13059-026-04309-2

Image Credits: AI Generated

DOI: 10.1186/s13059-026-04309-2

Keywords: membrane transporters, substrate specificity, TraSPIN, protein language models, geometric deep learning, graph learning, solute carriers, de-orphanization, protein-substrate interactions, functional annotation, Genome Biology, machine learning

But a stubborn handful of receptors has resisted every attempt at de-orphanization. TAS2R19, TAS2R42, TAS2R45, and TAS2R60 remain orphans, receptors with no known activating compounds, and their biological roles have therefore remained shrouded in uncertainty. At the same time, even among receptors classified as functional, genetic variation has produced variants labeled nonfunctional, sometimes on the basis of just one or a few amino acid substitutions. Two famous examples are TAS2R9 and TAS2R38. In TAS2R9, an alanine at position 187 yields a functional receptor, while a valine at the same position renders it unresponsive to known agonists such as the antibiotic ofloxacin. TAS2R38 is even more celebrated: its PAV variant makes people exquisitely sensitive to the bitter compounds phenylthiocarbamide and propylthiouracil, whereas the AVI variant is classed as nonfunctional and defines the classic taster-non-taster divide in human populations. Whether such variants are truly dead or have simply shifted their ligand preferences to unknown chemicals has been impossible to determine from sequence alone.

A new study by Praveen Kumar and Maik Behrens, published in Current Research in Food Science, tackles this puzzle from an unexpected angle. Instead of hunting for missing agonists, the researchers asked a more fundamental question: do these orphan and supposedly nonfunctional receptors even reach the cell surface, the location where a receptor must reside to detect compounds outside the cell? A receptor that fails to fold properly, or that gets stuck in the endoplasmic reticulum during biosynthesis, can never respond to any ligand, no matter how thorough the screening. Conversely, a receptor that traffics efficiently to the plasma membrane is structurally intact and remains a plausible candidate for future ligand discovery. The answer to this question provides a powerful clue about functionality that does not require knowing a single agonist.

To measure cell surface localization, the team employed a clever tagging strategy developed recently in their laboratory. Each receptor was engineered to carry an 11-amino-acid HiBiT sequence, a small peptide tag appended to the amino-terminal end of an sst3 export tag derived from the rat somatostatin receptor subtype 3, which is itself crucial for routing receptors to the plasma membrane. The HiBiT tag binds with high affinity to a complementary protein fragment called LgBiT, and together the two fragments generate luminescence. Because some of the commercial detection reagents are membrane-impermeable, they can only reach HiBiT tags exposed on the outside of living cells, allowing researchers to quantify exactly how much of each receptor sits at the cell surface. A second, lytic detection reagent breaks the cells open and reveals the total receptor population, permitting calculation of the fraction that successfully completed trafficking. The entire workflow runs on a fluorometric imaging plate reader, making it suitable for high-throughput analysis.

The first test cases were the taster and non-taster variants of TAS2R9 and TAS2R38. Functional assays in HEK293T-Gα16gust44 cells confirmed the expected behavior: the TAS2R38-PAV variant responded to propylthiouracil and the TAS2R9-187A variant responded to ofloxacin, while their non-taster counterparts remained silent. Importantly, adding the HiBiT tag did not abolish function, although the tagged TAS2R38 showed a slightly reduced response, a caveat the authors note openly. When the researchers then measured cell surface abundance using the luminescence assay, both functional and nonfunctional variants proved to be present at the plasma membrane. The non-taster TAS2R38-AVI variant reached the cell surface at a reduced fraction, about 6 percent of its total pool compared with 17 percent for the PAV variant, and the TAS2R9-187V variant showed only a modest reduction relative to its functional partner. Immunocytochemical staining of living, non-permeabilized cells with an anti-HiBiT antibody independently confirmed these surface-localization patterns.

These results carry significant implications for how scientists interpret receptor polymorphisms. For TAS2R9-187V, the data suggest that no gross defect in folding or routing explains its apparent nonfunctionality. Position 187 sits in the upper part of the fifth transmembrane domain, a region that could plausibly contribute to agonist binding. The authors therefore propose that rather than being a lost receptor, this variant may simply respond to a different, perhaps smaller, agonist that has not yet been tested. The bulkier valine residue could sterically reshape the ligand-binding pocket or restrict the conformational movements of the transmembrane helices needed for receptor activation. Similarly, for TAS2R38-AVI, the demonstration of structural integrity at the cell surface supports earlier speculation that this variant may have evolved altered ligand specificity, possibly responding to unknown compounds in the pulp of the Antidesma bunius fruit, which produces a peculiar bitter taste in individuals carrying the AVI haplotype.

The four orphan receptors yielded equally revealing results. All four could be detected at the cell surface by luminescence assays, but their trafficking efficiencies differed dramatically. TAS2R19 stood out as the star performer: 83 percent of its total receptor pool reached the plasma membrane, a proportion exceeding even that of TAS2R14, the human bitter receptor with the largest number of known agonists. TAS2R60 was by far the most abundantly produced receptor subtype, and although only 9 percent of it reached the surface, the absolute quantity there surpassed all TAS2Rs except TAS2R19. In sharp contrast, TAS2R42 and TAS2R45 showed negligible surface signals in immunocytochemical staining, with cell surface fractions of just 3 and 6 percent respectively. TAS2R42 is synthesized reasonably well but trafficked poorly, suggesting retention in the endoplasmic reticulum, while TAS2R45 appears to be produced at very low levels overall, raising the possibility that it may functionally represent an expressed pseudogene.

The contrasting profiles of the four orphans allow the authors to draw nuanced conclusions about why these receptors have resisted de-orphanization. For TAS2R19 and TAS2R60, biosynthesis and cell surface routing in heterologous cells clearly do not explain the absence of known agonists. Both receptors are closely related to already de-orphaned receptors in mice: the mouse receptor Tas2r135, an ortholog of TAS2R60, has 11 identified bitter agonists, and TAS2R19 is highly homologous to TAS2R20, which has 10 known agonists. Problems with coupling to the calcium signaling cascade used in the assay system also seem unlikely, given that far more distantly related receptors, including one from cartilaginous fish, have been successfully screened in the same cellular platform. The most probable explanation for these two receptors is simply that the right bitter compounds have not yet been tested, pointing toward unexplored chemical territory.

That territory, the authors argue, is vast and seriously under-sampled. Most bitter agonist discovery has focused on synthetic compounds and molecules from flowering plants, while bitter substances from animals, mushrooms and other fungi, prokaryotic organisms, and plants belonging to older lineages of the plant kingdom have barely been investigated. Recent work identifying bitter compounds from mushrooms, such as those isolated from the bracket fungus Amaropostia stiptica, illustrates the kind of neglected sources that could harbor activators for the remaining orphans. Closing this knowledge gap, the study suggests, will require intensified bioprospecting efforts guided by the growing recognition that bitter taste receptors do far more than guard the tongue. They are expressed in the gut, airways, and other tissues, where they influence hormone secretion, glucose homeostasis, and possibly innate defense, making every de-orphanized receptor a potential target for nutrition and medicine.

The HiBiT-based localization assay itself may prove to be the study’s most lasting contribution. By providing a rapid, quantitative, high-throughput readout of receptor biosynthetic integrity that requires no knowledge of agonists, the method offers a rational triage tool for the bitter taste receptor field. Receptors that traffic well, like TAS2R19 and TAS2R60, become priority targets for expanded ligand screening, while those with severe trafficking defects, like TAS2R42 and TAS2R45, may need engineered chaperones, modified expression systems, or tissue-specific organoid models to reveal their secrets. As the authors note, genetically engineered organoids derived from taste tissue or the intestine could eventually provide more natural cellular environments for such studies, overcoming limitations of immortalized cell lines that express numerous receptors endogenously. For now, the message is clear: four orphan receptors and two supposedly dead receptor variants are very much alive at the cell surface, waiting for the right molecule to wake them.

Subject of Research: Cell surface localization of orphan and nonfunctional human bitter taste receptors

Article Title: Cell surface localization of receptors considered nonfunctional and orphan bitter taste receptors

Article References: Kumar, P., & Behrens, M. (2026). Cell surface localization of receptors considered nonfunctional and orphan bitter taste receptors. Current Research in Food Science, 13, Article 101551. https://doi.org/10.1016/j.crfs.2026.101551

Image Credits: AI Generated

DOI: 10.1016/j.crfs.2026.101551

Keywords: bitter taste receptors, TAS2R, orphan receptors, HiBiT tag, cell surface localization, G protein-coupled receptors, TAS2R38, TAS2R9, de-orphanization, taste perception, receptor trafficking, food science