Maximum Entropy Is Helping Scientists Predict How Mutations Change Enzymes and Drive Drug Resistance

For decades, enzymes have fascinated scientists because they sit at the heart of almost every biological process. From metabolism to viral replication, these molecular machines determine how life functions at a chemical level. One scientist who has spent his entire career trying to understand enzymes is Arieh Warshel, a Distinguished Professor of Chemistry at the University of Southern California and a 2013 Nobel Prize laureate.

Warshel is best known for pioneering computer simulations that explain how enzymes speed up chemical reactions. But around five years ago, he ran into a wall. His traditional physics-based simulations, despite being incredibly sophisticated, were no longer sufficient to answer the questions that mattered most—especially how mutations change enzyme behavior and how those changes allow viruses to evade drug treatments.

Instead of giving up, Warshel took a different route. He turned to artificial intelligence and statistical physics, focusing on a concept known as maximum entropy. What followed has opened a new way of thinking about enzyme function, disease-causing mutations, and drug resistance.

Why Traditional Simulations Fell Short

For years, Warshel relied on detailed molecular simulations to model enzymes atom by atom. These methods are powerful but also computationally expensive and slow. When studying mutations—especially in viruses—the number of possible changes becomes overwhelming. Testing mutations one by one simply does not scale.

This limitation became especially clear when studying viruses such as HIV and hepatitis C virus (HCV). These pathogens mutate rapidly, and even small changes in enzyme structure can dramatically alter how well drugs work. Predicting which mutations will emerge next is one of the hardest problems in modern biomedical research.

Warshel realized that instead of simulating every physical detail, he might be able to extract meaningful predictions from statistics alone.

The Discovery of a Powerful Pattern

Warshel and his team began exploring whether enzyme activity could be linked to maximum entropy, a statistical measure that describes how many different configurations a system can adopt while still satisfying certain constraints.

Across several studies, they found something striking: enzyme activity consistently correlated with maximum entropy values. In simple terms, enzymes that had higher entropy—meaning more statistically favorable configurations—tended to function faster or more efficiently.

This discovery suggested that enzyme behavior could be predicted without simulating every atomic interaction. Instead, a purely computational and statistical approach could reveal how mutations influence function.

That insight opened the door to a much bigger question: could maximum entropy also help predict drug resistance?

Applying Maximum Entropy to Viral Drug Resistance

Drug resistance occurs when mutations allow viruses to continue replicating despite treatment. Warshel had tackled this problem before. In 2008, he built models to predict how HIV might escape drug pressure. While those models had some success, they were slow and could not reliably anticipate future mutations.

With maximum entropy, the approach was different. Instead of testing mutations individually, the model evaluates the entire mutational landscape at once, considering both how strong a mutation is and how likely it is to occur.

HIV was the first test case, partly because of its importance and partly because Stanford University maintains a massive database of HIV mutations observed in patients undergoing treatment.

The results were mixed. Maximum entropy did correlate with patterns of drug resistance in HIV. However, a much simpler measure—the raw number of accumulated mutations—showed a similar correlation. HIV mutates in nearly every possible way, making it extraordinarily difficult to predict its next move.

In practical terms, the method could compare drugs already tested in patients, but it did not outperform what clinicians could already observe.

A Breakthrough with Hepatitis C Virus

The real breakthrough came when the team shifted focus to viruses with more constrained evolutionary pathways. One of the first was hepatitis C virus (HCV).

Unlike HIV, HCV cannot mutate freely without harming itself. In this system, maximum entropy aligned much more clearly with the mutations that actually emerged under drug pressure. This meant researchers could potentially forecast future resistance mutations, rather than simply explaining them after the fact.

Warshel described this as playing chess with a virus—anticipating not just what moves are possible, but which ones are most likely. The findings from this work were recently published in the Proceedings of the National Academy of Sciences (PNAS).

Beyond Viruses: Insights into Human Disease

Drug resistance is only part of the story. As Warshel’s team applied maximum entropy to other biological systems, they found it worked remarkably well beyond virology.

One major area of success involves myosin, a molecular motor protein essential for muscle contraction, heart function, and hearing. Mutations in myosin proteins are known to cause inherited forms of deafness and serious heart conditions.

In these systems, maximum entropy showed strong and reproducible correlations with disease-causing mutations and with how drugs interact with the proteins. In many cases, the method was faster, easier, and more accurate than the brute-force simulations Warshel had used for decades.

A new paper from his group, currently under review, reports that maximum entropy can even predict how efficiently myosin “walks”—a term used to describe how it moves along cellular structures when altered by specific mutations.

What Is Maximum Entropy and Why Does It Matter?

At its core, maximum entropy is a principle from statistical physics. It describes the most unbiased probability distribution that fits known data while assuming as little as possible beyond that data.

In biology, this means researchers can analyze mutation frequencies and sequence data to infer which configurations are most likely to occur. When those predictions align with real-world outcomes—such as enzyme activity or drug resistance—the method becomes a powerful tool.

The appeal of maximum entropy lies in its efficiency. Instead of modeling every physical interaction, scientists can focus on patterns that emerge from large datasets, making it especially well suited for modern biology, where data is abundant but time and computing power are limited.

The Bigger Picture for Drug Design and Medicine

Although HIV remains too unpredictable for current maximum entropy models to fully tame, the broader implications are significant. Many pathogens, enzymes, and disease-related proteins operate within more restricted evolutionary limits, making them ideal candidates for this approach.

By identifying which mutations are both effective and likely, researchers could design drugs that are harder for viruses or diseased proteins to escape. Beyond infectious diseases, this method may reshape how scientists study genetic disorders, enzyme engineering, and protein-based therapies.

For Warshel, the shift from traditional simulations to statistical modeling represents not a departure from his life’s work, but an evolution of it. Maximum entropy has become a new lens—one that reveals how biological systems respond to mutation with surprising clarity.

Research Reference:
https://doi.org/10.1073/pnas.2517715122