What Happened in the World of Bioinformatics in 2025?

In recent years, artificial intelligence has evolved from being merely an auxiliary tool in biology to becoming a direct actor that shapes the discovery process itself. From protein binding to genome design, from gene editing experiments to disease prediction, a common theme emerges across many fields: fewer trials, more predictions.

In this blog content, I will discuss how artificial intelligence is creating a new paradigm in bioinformatics and genetic engineering by examining four different approaches announced recently.

1. Boltz-2: Binding Prediction Without Simulation

Understanding how strongly a drug binds to its target protein traditionally relies on lengthy and costly molecular simulations. Boltz-2 offers an approach that directly challenges this assumption.

This open-source model can predict binding affinity in early-stage drug screening up to a thousand times faster than classical methods. Moreover, this speed increase is achieved without compromising accuracy.

What makes **Boltz-2 **truly significant is that it offers more than just a technical improvement. The model represents a new research culture that pushes simulations into the background by combining the physics-based intuitive learning approach pioneered by AlphaFold with millions of biochemical experimental data points.

The potential outcomes of this approach are clear:

• Fewer misses
• Lower experimental costs
• Faster feedback loops

And perhaps most importantly, the possibility for small teams to tackle questions that previously only large industrial laboratories dared to address.

2. Functional Intervention with Protein Fragments: FragFold

Protein-protein interactions form the basis of cellular life. However, how these interactions are regulated at the molecular level remains largely unknown.

FragFold offers a compelling perspective on this issue. The model can predict whether protein fragments, consisting of short amino acid sequences, can mimic natural interactions by binding to the interfaces of target proteins.

Recent findings show that these small protein fragments have a much greater functional potential than expected. When the correct binding modes are captured, protein functions can be altered or specific interactions can be deliberately disrupted.

This approach has a wide range of applications, from understanding protein-ligand interactions to synthetic protein design.

3. From Reading to Designing: Evo 2 and Generative Genomics

AlphaFold demonstrated that structure is predictable in biology. Evo 2 goes one step further, suggesting that the genome can become a designable object.

Trained on billions of parameters, this model can analyze DNA, RNA, and proteins together across trillions of nucleotides. Protein structure prediction, new molecule discovery, and evaluation of the functional effects of mutations are just a few of its capabilities.

The significance of Evo 2 lies not so much in solving individual tasks, but in its potential to transform biological design into a systematic and scalable process. Traditionally slow, unpredictable, and largely manual processes are becoming increasingly computable.

4. Artificial Intelligence Assistant in Experiment Design: CRISPR-GPT

Gene editing experiments rely on practical intuition as much as theoretical knowledge. CRISPR-GPT is a system that aims to digitize precisely this intuition.

Thanks to its multi-faceted architecture, the model does more than just answer questions. It plans experiment steps, anticipates potential errors, and provides guidance tailored to the user's level.

The system's foundation in thousands of discussions among scientists over more than a decade sets it apart from a classic knowledge base. Its learning of the thought processes specific to laboratory practice creates a significant time advantage, especially in preclinical stages.

The long-term goal is clear: for these systems, which begin under human control, to evolve into a more autonomous structure in experimental biology by merging with robotic platforms.

2025 Summary and Outlook for 2026

While these four examples appear to address different problems at first glance, they point to a common thread that will become increasingly clear throughout 2025. Artificial intelligence is no longer merely a tool that accelerates processes in biology. It has reached an entirely new level. It has become a partner that reduces the burden of trial and error and shares the researcher's intuitive decisions. And I'm sure there's more to come. It could be good or bad :/

In my opinion, perhaps the most distinctive feature of 2025 was this: biology began to become “more strategic” rather than “faster.” Predicting protein binding without simulation, functional intervention with protein fragments, making the genome designable with generative models, and delegating experiment planning to artificial intelligence. All of these were different facets of the same transformation.

This year showed us, in short, that the real bottleneck is no longer data or computing power, but knowing which questions to ask. As artificial intelligence reduces the cost of searching for answers, the value of asking good questions increases even more.

This is precisely where the promise lies as we enter 2026. Next year, we can expect these systems to gain more autonomy, integrate more tightly with laboratory infrastructure, and blur the lines between experimental biology and software even further. A research environment where small teams can formulate big hypotheses and make bolder discoveries with fewer resources seems more possible than ever.

Perhaps the question we should be asking now is not: How much can artificial intelligence be integrated into biology?

The real question is: By 2026, how prepared will we be to think about biology alongside artificial intelligence?

A Few Predictions for 2026

By 2026, this transformation is likely to start producing more concrete results. Some trends are already becoming apparent.

Modeling before experimentation will become the default approach (it's somewhat like this now, but it will increase). Most hypotheses will be filtered out by AI-supported systems before entering the wet lab. Experiments will be conducted for verification rather than discovery. Biological design will evolve into a software-like process. It will be associated with concepts such as protein, gene, and circuit design, versioning, rollback, and rapid iteration.

Small and agile teams will become competitors to large laboratories. Computation-based predictions will neutralize some of the advantages of capital and infrastructure. The role of the researcher will change. Less manual optimization and more problem framing and strategic decision-making will come to the fore.

The boundary between biology and software will become even more blurred. Writing code will become as fundamental a skill as designing experiments, while biological intuition will increasingly influence model development processes. Of course, I'm not talking about utopian predictions. These are already the things with the highest probability of happening. So how prepared are we for this transformation? That's what matters...