Decoding the metabolic orchestra

Every cell in the human body, along with the trillions of microbes that live on and inside us, runs like a minifactory. Raw materials flow in, enzymes reshape them, and finished products are used locally or shipped elsewhere. This seamless choreography of chemical reactions is metabolism – a molecular symphony performed simultaneously by trillions of cells.

Understanding the orchestration of this music has never been simple. Knowing how to silence a specific disease-causing note without disrupting the rest of the symphony is incredibly harder. But advances in artificial intelligence, computer modelling and simulation, and digital twins elevate scientists to music conductors, empowering them to reveal not only what is happening inside the body, but also where and how to intervene at a molecular level.

A striking example comes from researchers at the University of Michigan, who recently applied metabolic modelling of glioma, one of the deadliest forms of brain tumor. Gliomas account for roughly 80 percent of malignant brain tumours and are notoriously unpredictable. Two tumours that look identical on a medical image may behave very differently at the molecular level. The same therapy can have positive effects on one, and have no impact on another.

One promising tactic is to starve tumours by disrupting their metabolism. Some gliomas depend on specific amino acids to make guanosine, a building block of DNA. Block that pathway and the tumour has a hard time to replicate and spread. Doctors can attempt this chemically, using drugs such as mycophenolate mofetil, or nutritionally, by restricting certain amino acids in a patient’s diet.

Yet, tumours are resourceful. Some synthesize their own amino acids, rendering diet-based restrictions useless, others can also import guanosine from the surroundings, bypassing drugs. Without knowing which strategy a tumour will resist, precious time can be lost in trial and error.

To cut through that uncertainty, the Michigan team built a machine learning-based digital metabolic twin of gliomas by combining blood metabolite levels, tumour genetic profiles and metabolic data. They help reconstruct personalised maps of how each tumour processes nutrients. 

The computer model was trained on synthetic patient data grounded in established principles of biochemistry and validated against measurements from real glioma patients. Subsequently, the model was tested in mice. The digital twin predicted which tumours would respond positively to the dietary restriction, and which would not. When the diet restriction was applied to mice with glioma, tumour growth slowed specifically within those animals that the computer model  had identified as likely responders. In all other animals, the cancer continued unimpeded.

The same approach was applied to the drug mycophenolate mofetil that impedes specific amino acid production. The computer model helped identify those tumours that could sidestep the drug’s mechanism and continue to absorb necessary amino acids from the environment, for replication. Mouse experiments also confirmed the predictions.

The implication is profound. Instead of weeks spent on ineffective treatments, a digital twin could identify the best strategy from the start.

The approach extends far beyond cancer. Some of the most surprising applications of metabolic modelling may act far away from the target organ. For instance, they can focus on the gut microbiome to exert their effect on the brain.

At the University of Galway, VPH member Prof. Ines Thiele and her colleagues at the Digital Metabolic Twin Centre applied whole-body metabolic models that integrate organs, sex differences and the gut microbiome into a single computational system. Their simulations can trace how a bacterium in the intestine might alter a molecule detected in the blood, or influence disease risk.

In Parkinson’s disease, this approach has produced striking findings. Analysing gut microbiome data from nearly 500 Parkinson’s patients and 234 healthy controls, Thiele’s team predicted that in Parkinson’s patients, their microbiome has a reduced capacity to deliver several key molecules to the bloodstream, including the amino acid L-leucine, as well as butyrate, nicotinic acid and myristic acid.

Each shortfall was traced to a specific bacteria. Roseburia intestinalis, for instance, appeared to produce less L-leucine in Parkinson’s patients, while Methanobrevibacter smithii consumed more of it. Faecalibacterium prausnitzii, a broadly recognised marker of gut health, was linked to reduced butyrate production. These findings do not merely describe an association, they propose a mechanism, a causal chain running from changes in microbial populations, through altered blood chemistry, to neurological symptoms.

A parallel investigation applied the same approach to Alzheimer’s disease. Based on data from over a thousand participants in the Rotterdam study, a long-running cohort of healthy middle-aged and elderly Europeans, Thiele’s group built a personalised whole-body computational model incorporating their microbiome composition.

The computational model connected gut microbial metabolism to several of Alzheimer’s key risk factors such as age, sex, cognitive performance, and the APOE gene, a variant that strongly influences a person’s risk of developing the disease. With aging, the models predicted an increased microbial capacity to supply the body with L-arginine, an amino acid. The byproduct of arginine breakdown  may help suppress the build-up of amyloid-beta, the protein that accumulates in Alzheimer’s brains.

Likewise, bile acids also emerged as another thread to leverage. Secondary bile acids like deoxycholate and lithocholate, produced when gut bacteria chemically transform bile acids made in the liver, showed a striking link to cognition. Higher levels of deoxycholate and lithocholate, both toxic to cells, were associated with lower cognitive scores. Lower deoxycholate production, by contrast, correlated with the APOE E2, a genetic variant of the gene APOE associated with reduced Alzheimer’s risk. Metabolomics data from the same participants corroborated the computer model’s predictions, grounding the computational findings to quantifying biological changes.

Notably, no single bacterial species drove these complex and sophisticated effects. The correlations were the product of countless interactions between microbes and their host that no standard experiment could easily unravel. Just like the master conductors orchestrating hundreds of musicians play complex music compositions.

This is exactly where metabolic modeling proves its worth. It captures the single chords and instruments of the symphony of life to radically improve how we understand and treat diseases.

Further readings:

• Digital twins for in vivo metabolic flux estimations in patients with brain cancer, Cell metabolism, 2026
• Metabolic modeling links gut microbiota to metabolic markers of Parkinson’s disease, Gut Microbes, 2025
• In silico metabolic modelling links microbiome-derived metabolites to risk factors of Alzheimer’s disease, Gut Microbes Report, 2024
• Human Systems Biology and Metabolic Modelling: A Review—From Disease Metabolism to Precision Medicine, BioMed Research International, 2019