Predicting GPCR Function: Inside the Carlsson Lab’s Modeling Toolbox

Dr. GPCR Podcast
2 days ago
4 min read

AlphaFold to Assay: Predictions That Last

Watch Episode 175

If your model can’t predict the future of GPCR drug discovery, why build it at all?

For decades, GPCRs have been the cornerstone of pharmacology. Nearly one-third of all approved drugs target them. Yet even with the structural biology revolution, one stubborn hurdle persists: prediction.

Can we leverage structural and computational insights not just to explain receptor–ligand interactions after the fact, but to forecast outcomes and design ligands with new properties?

That’s the mission of Dr. Jens Carlsson, Professor of Computational Biochemistry at Uppsala University.

His lab sits at the crossroads of structure-based modeling, computational chemistry, and drug discovery—rethinking how simulations, docking, and machine learning can transform GPCR research from descriptive science into predictive decision-making.

From Static Structures to Testable Hypotheses

Carlsson’s philosophy is simple but radical: modeling should generate hypotheses, not just rationalize data.

Projects in his group often begin with either:

A new GPCR structure ripe for exploration, or
A pharmacological challenge such as differentiating agonists from antagonists, predicting biased agonism, or achieving receptor subtype selectivity.

To tackle these, the lab employs molecular docking, molecular dynamics simulations, and ligand-based screening. Vast chemical libraries—sometimes billions of molecules—are sifted computationally to highlight potential hits.

When existing compounds are exhausted, bespoke ligands are designed and synthesized in-house. That medicinal chemistry capacity is unusual for a modeling lab and provides a critical bridge between prediction and validation.

Every computational hypothesis moves downstream into experimental assays via close collaborations. For Carlsson, a prediction is only valid if it survives wet-lab testing.

The “Predict, Not Explain” Ethos

What sets the Carlsson lab apart is its internal rule: results must be actionable.

Instead of rehashing how a known ligand binds to a known receptor, they aim to produce predictions that guide experimental choices—whether in target prioritization, compound selection, or mechanistic exploration.

This predictive framing raises the bar. Common questions in the group include:

Can we forecast ligand efficacy or selectivity?

Can we design compounds that trigger biased signaling?

But part of their rigor lies in restraint. If a method can’t resolve subtle potency differences, the team prefers to acknowledge that limitation rather than overpromise.

That scientific humility has earned credibility with experimental collaborators, where skepticism of computational work is common.

GPCR Modeling in Context: Then and Now

When Carlsson entered the GPCR field in the early 2000s, structural data was scarce. His early Adenosine 2A receptor work, in collaboration with Ken Jacobson, helped establish virtual screening as a viable tool for GPCR ligand discovery.

Fast forward to today: dozens of GPCR structures exist, many in multiple conformations or bound to G proteins and arrestins. Yet functional selectivity, allosteric modulation, and biased agonism remain major challenges.

Carlsson’s group integrates modeling with mechanistic pharmacology.

For dopamine receptor studies, they built predictive models before experimental structures were available. When crystal structures were finally solved, the models aligned strikingly well—a rare and powerful validation.

AlphaFold: A Game-Changer with Caveats

One of the newest tools in the lab’s arsenal is AlphaFold, DeepMind’s AI-powered protein structure prediction platform.

Carlsson’s team has already published cases where AlphaFold-derived GPCR models provided accurate scaffolds for ligand docking—enabling progress in systems without experimental structures. In one instance, an AlphaFold prediction was so strong the group pivoted its entire workflow around it.

But AlphaFold is not a panacea. It struggles with ligand-bound conformations, GPCR–G protein complexes, and receptor–peptide interactions.

The lab has witnessed both spectacular accuracy and puzzling failures.

Carlsson’s takeaway: AlphaFold is a powerful starting point, but domain expertise and experimental grounding remain indispensable.

Translational Impact for Drug Discovery

While the Carlsson lab is rooted in academia, its impact extends into biotech and pharma.

Carlsson also runs a consulting arm that advises drug discovery teams on GPCR modeling strategy. Notably, this isn’t contract computational chemistry—it’s strategic guidance on target selection, ligand feasibility, and prioritization.

This dual perspective makes Carlsson a trusted partner for industry teams navigating early-stage pipelines, where skepticism toward in silico hits is high. By focusing on translationally relevant predictions, the lab helps bridge the gap between computation and experiment.

For drug discovery executives, the message is clear: predictive modeling can shorten timelines, reduce costs, and expand druggable space—if it’s done with rigor and honesty.

Why Predictive GPCR Modeling Matters Now

The convergence of machine learning, improved GPCR structures, and scalable experimental assays is shrinking the gap between modeling and pharmacology. But success depends on researchers who can manage expectations, build trust, and integrate across disciplines.

Carlsson’s group does not aim for perfect predictions.

Instead, it aims for useful, testable predictions that shape the next experiment.

That shift—from describing the past to forecasting the future—is what makes predictive GPCR modeling transformative.

For scientists, this is an intellectual challenge: demand more from your models. For industry leaders, this is an operational opportunity: integrate prediction into decision-making.

Modeling That Moves the Field Forward

The Carlsson lab exemplifies how computational biochemistry can meaningfully impact real-world pharmacology. With rigor, transparency, and cross-disciplinary collaboration, they’re pushing GPCR science toward a predictive era.

The next frontier in GPCR drug discovery isn’t more structures—it’s smarter models. Are you demanding enough from yours?

Want to go deeper into GPCR modeling?

Dr. Jens Carlsson breaks down the tools, challenges, and mindset behind building models that don’t just explain—they predict.

🎧 Listen to the full episode of the Dr. GPCR Podcast