While AlphaFold revolutionized how scientists predict protein shapes, a critical gap remains: predicting how proteins interact with each other and with essential metals like copper and zinc. A Texas A&M undergraduate researcher is now filling that gap with a machine learning tool that could unlock new drug targets for diseases ranging from copper deficiency disorders to cancer.

Dimitris Kalafatis, a junior biochemistry student at Texas A&M University, has been named a 2026 Barry Goldwater Scholar for developing a computational program that maps protein-protein and protein-metal interactions across thousands of human proteins simultaneously. The award, given to fewer than 450 students annually from over 5,000 nominees nationwide, recognizes his work as among the most prestigious undergraduate research honors in the nation.

What Can AlphaFold Predict, and What Can't It?

Google DeepMind's AlphaFold earned a Nobel Prize in Chemistry in 2024 for solving the protein folding problem, a challenge that had stumped scientists for over 50 years. The system can predict the three-dimensional structure of a protein from its amino acid sequence in minutes, a task that once required years of laboratory work and hundreds of thousands of dollars. Before AlphaFold, researchers had experimentally determined roughly 190,000 protein structures over six decades. The AI system extended that coverage to over 200 million proteins, essentially cataloguing all known proteins to science.

But AlphaFold has a critical limitation: it cannot accurately predict how proteins interact with each other or bind to metals. This matters enormously for drug discovery, because most drugs work by binding to target proteins, and many proteins require metals like copper or zinc to function properly. Without understanding these interactions, researchers miss potential drug targets and therapeutic pathways.

How Is Kalafatis Filling the Gap?

Working in the laboratory of Vishal Gohil, a professor and Chancellor EDGES Fellow in the Department of Biochemistry and Biophysics, Kalafatis teamed up with a senior graduate student, Abhinav Swaminathan, to develop an AI-based prediction tool specifically designed to map which proteins bind to metals and how. The tool uses machine learning to analyze patterns across thousands of proteins at once, leveraging Texas A&M's High Performance Research Computing Facility to run predictions at scale.

The inspiration came directly from the lab's own discovery: elesclomol, a copper-transporting drug now used to treat Menkes disease, a rare copper deficiency disorder that causes neurological damage in children. Kalafatis realized that if his team could systematically predict protein-metal interactions across the entire human proteome, they could identify dozens or hundreds of similar drug targets that had previously gone unnoticed.

"I found a project that was really cool and really mathy, two things I love," Kalafatis said. "I decided, oh, actually I think this is what I want to do."

Dimitris Kalafatis, Undergraduate Researcher, Texas A&M University

Why Does This Matter for Drug Discovery?

The human body contains roughly 1,000 proteins in mitochondria alone, the cellular structures that generate biological energy. Scientists do not fully understand what many of these proteins do, and this knowledge gap has hindered their ability to understand how mitochondria malfunction and lead to disease. Kalafatis thinks of proteins as workers inside every cell, each with a specific job. But most cannot do that job alone; many proteins interact with each other, and some need metals to function. Without them, things go wrong at the cellular level, sometimes causing serious diseases.

By systematically mapping these interactions, Kalafatis' tool could point researchers toward previously unknown drug targets. This is particularly valuable for rare diseases, where traditional drug discovery approaches often fail because the biological mechanisms are poorly understood. The ability to predict protein-metal interactions at scale could make therapies for rare diseases more accessible and faster to develop.

How to Advance Protein-Based Drug Discovery Beyond AlphaFold

• Combine Structure Prediction with Interaction Modeling: Use AlphaFold to predict individual protein shapes, then apply specialized machine learning tools to predict how those proteins interact with each other and with cofactors like metals.
• Leverage High-Performance Computing: Scale prediction tools across thousands of proteins simultaneously using supercomputing facilities, which can identify novel interactions that would be impossible to discover through traditional laboratory methods.
• Validate Predictions in the Lab: Use computational predictions to prioritize which protein-metal interactions to test experimentally, reducing the time and cost of traditional drug discovery pipelines.
• Focus on Rare Disease Targets: Apply these tools to diseases where protein function is poorly understood, since computational predictions can fill knowledge gaps that slow down traditional research.

Kalafatis' work represents a broader trend in computational biology: as AlphaFold solved the structure prediction problem, researchers are now tackling the next frontier, understanding how proteins work together in living cells. His research demonstrates that the most impactful discoveries often come not from replacing traditional biology with AI, but from using AI to answer questions that traditional methods cannot address efficiently.

The Goldwater Scholarship, established by Congress in 1986 to honor U.S. Senator Barry Goldwater, has a track record of identifying researchers who go on to earn Rhodes Scholarships, Churchill Scholarships, and National Science Foundation Graduate Research Fellowships. Kalafatis plans to apply to dual medical and research doctoral programs this fall, with the long-term goal of using protein folding research to develop new drugs and make science-based medicine more accessible worldwide.