Machine Learning Finds Mango Compound That Targets Diabetes Protein

A compound found in mangoes may have a far more precise role in fighting diabetes than scientists previously understood, according to a new study published in the journal 3 Biotech. An international team of researchers from India, Mauritius, and the Maldives combined machine learning, network pharmacology, molecular dynamics simulations, and laboratory experiments to identify the transforming growth factor beta receptor 1, known as TGFBR1, as a previously undefined molecular target of mangiferin, a naturally occurring C-glucosyl xanthone extracted from Mangifera indica, the common mango tree. The work goes further than target identification alone: the researchers also evaluated a combination of mangiferin with atorvastatin, a widely prescribed statin drug, and found that the pairing improved glucose tolerance, lowered blood lipids, and preserved pancreatic tissue in rats fed a high-fat diet. The findings offer a mechanistic explanation for mangiferin’s long-observed antidiabetic activity and suggest a possible route toward combination therapies built on existing medicines.

The central problem the researchers set out to solve was a familiar one in pharmacognosy, the study of medicines derived from natural sources. Mangiferin has been repeatedly associated with antidiabetic effects in the scientific literature, yet the specific protein through which it exerts this activity had never been conclusively defined. Traditional approaches to target identification can take years, requiring extensive biochemical screening of candidate proteins one at a time. The team instead turned to supervised machine learning, training computational models on a curated collection of known TGFBR1 inhibitors obtained from BindingDB, a public database of protein-small molecule binding data. By learning the molecular features that distinguish compounds capable of inhibiting this receptor, the models could then predict whether previously untested phytochemicals would behave similarly.

Several supervised learning algorithms were trained and compared, and the standout performer was a Random Forest model, an ensemble method that builds many decision trees during training and averages their predictions to reduce error and overfitting. The Random Forest classifier achieved an accuracy of 94.68 percent with a Cohen’s kappa coefficient of 0.8981, a statistic that measures agreement between predicted and actual classifications while correcting for chance. Crucially, the model also demonstrated robust performance on external validation, reaching 79.35 percent accuracy on 868 compounds it had never encountered during training. External validation of this kind is a critical safeguard in computational drug discovery, because models that perform well only on their training data often fail when confronted with chemically novel structures in the real world.

With a validated predictive model in hand, the researchers screened a library of 545 phytochemicals derived from Mangifera indica. Five compounds emerged as high-confidence predicted inhibitors of TGFBR1: mangiferin itself, along with fisetin, isochinomin, mangiferolic acid, and astragalin. The identification of mangiferin at the top of this list provided computational support for the hypothesis that this abundant mango xanthone could interact with the TGF-beta signaling pathway, a cascade involved in cell growth, immune regulation, and metabolic processes that has been increasingly implicated in the tissue damage associated with diabetes and its complications.

To place these individual predictions in a broader biological context, the team employed network pharmacology, an approach that maps the interactions among drugs, targets, diseases, and signaling pathways to reveal the most influential nodes in a biological system. Their analysis identified TGF-beta as the most central protein in the interaction network, with a cumulative score of 350, reinforcing the idea that the TGF-beta pathway sits at the heart of mangiferin’s antidiabetic mechanism. Molecular docking studies then examined how mangiferin fits into the binding site of TGFBR1 at atomic resolution. The compound was found to form stable hydrogen bonds with three amino acid residues in the receptor’s binding pocket, serine 287, isoleucine 211, and glutamic acid 208, supplemented by hydrophobic contacts that help anchor the ligand in place.

Docking provides a static snapshot, so the researchers next turned to molecular dynamics simulations to test whether the predicted binding pose would hold up under realistic, thermally fluctuating conditions. Over a simulation lasting 100 nanoseconds, the mangiferin-TGFBR1 complex remained structurally stable, with root-mean-square deviation values between 1.50 and 1.75 angstroms, a range indicating that the protein-ligand complex did not drift apart or undergo disruptive conformational changes. Ligand-residue contact occupancy exceeded 85 percent, meaning that for the vast majority of the simulation the key interactions between mangiferin and the binding site residues were maintained. This level of dynamic stability is an encouraging sign that the predicted interaction reflects genuine physical binding rather than a fleeting computational artifact.

The team also assessed whether mangiferin possesses the chemical properties expected of a viable drug candidate. ADMET profiling, which evaluates absorption, distribution, metabolism, excretion, and toxicity, showed favorable drug-likeness, including compliance with Lipinski’s rule of five, a widely used heuristic for oral bioavailability, and low predicted toxicity. One limitation emerged: mangiferin’s topological polar surface area was calculated at 201.28 square angstroms, a value suggesting moderate permeability limitations that could affect how efficiently the compound crosses biological membranes. Complementary density functional theory calculations yielded a HOMO-LUMO gap of 2.85937 electron volts, an electronic property consistent with antioxidant activity, which is relevant given the role of oxidative stress in diabetic pathology.

The computational predictions were then subjected to experimental scrutiny in cultured cells. When atorvastatin was combined with mangiferin, the treatment increased apoptosis through the mitochondrial pathway, a form of programmed cell death characterized by a specific molecular sequence. The researchers observed downregulation of the anti-apoptotic protein Bcl-xL, upregulation of the pro-apoptotic protein BAK, release of cytochrome c from mitochondria into the cytoplasm, and activation of caspase-7 and caspase-8, the enzymatic executioners of the apoptotic program. Alongside these effects, the combination enhanced insulin sensitivity, demonstrated by increased glucose uptake in a 2-NBDG assay, a technique that uses a fluorescent glucose analog to track sugar transport into cells, and by elevated expression of the glucose transporter GLUT-4 and the phosphorylated insulin receptor substrate pIRS-1 at tyrosine position 896, both key markers of active insulin signaling.

The most clinically suggestive results came from animal studies. Rats fed a high-fat diet to induce metabolic dysfunction were treated with atorvastatin plus mangiferin for 24 weeks. Compared with untreated high-fat diet controls, the treated animals showed significantly improved glucose tolerance and reduced levels of HbA1c, the long-term glycemic marker reflecting average blood sugar over weeks to months, as well as lower total cholesterol, triglycerides, and random blood sugar measurements. Histological examination revealed preserved pancreatic architecture, indicating structural protection of the insulin-producing tissue, and muscle tissue from treated animals displayed increased GLUT-4 expression, consistent with enhanced peripheral glucose disposal. Together, these results suggest that the combination acts on multiple fronts simultaneously, improving lipid profiles through the statin component while the mango-derived compound contributes glycemic control through its newly characterized target.

The study, led by Raghuvir Keni, Farmiza Begum, Niraja Ranadive, Pawan G. Nayak, and Nandakumar Krishnadas of Manipal College of Pharmaceutical Sciences, together with collaborators at institutions including JSS Academy of Higher Education and Research in Mauritius and Vels Institute of Science, Technology and Advanced Studies in Chennai, represents a template for how artificial intelligence can compress the early stages of drug discovery. Rather than years of trial-and-error biochemical screening, a trained machine learning model evaluated hundreds of natural compounds in silico, flagged the most promising candidates, and directed experimental resources toward the hypotheses most likely to succeed. The authors caution that their findings, while spanning computation, cell culture, and animal models, remain preclinical, and the moderate permeability limitation identified in the ADMET analysis highlights a formulation challenge that would need to be addressed before any human application. Nevertheless, by naming TGFBR1 as a plausible molecular target of mangiferin and demonstrating that a mangiferin-atorvastatin combination improves metabolic parameters in a disease-relevant animal model, the research provides a concrete mechanistic foundation for a therapy concept that pairs a common fruit-derived molecule with one of the world’s most widely used prescription drugs. As diabetes continues to affect hundreds of millions of people worldwide according to the International Diabetes Federation, such repurposing strategies, grounded in computational prediction and validated step by step in the laboratory, offer a potentially faster and cheaper path to new treatment options than de novo drug design.

Subject of Research: Machine learning-driven identification of TGFBR1 as a molecular target of mangiferin from Mangifera indica and evaluation of a mangiferin-atorvastatin combination for antidiabetic therapy

Article Title: Machine learning-driven discovery of TGFBR1 inhibitors from Mangifera indica for Atorvastatin-based antidiabetic therapy

Article References: Keni, R., Begum, F., Ranadive, N., Nayak, P. G., Murugesan, V., Pradeep, K., Deshpande, M., Krishnadas, N., & Theivendren, P. (2026). Machine learning-driven discovery of TGFBR1 inhibitors from Mangifera indica for Atorvastatin-based antidiabetic therapy. 3 Biotech, 16(11), Article 458. https://doi.org/10.1007/s13205-026-05088-0

Image Credits: AI Generated

DOI: 10.1007/s13205-026-05088-0

Keywords: TGFBR1, mangiferin, Mangifera indica, machine learning, atorvastatin, diabetes, network pharmacology, molecular dynamics, drug discovery, phytochemicals, insulin sensitivity, Random Forest