For ten years, Mark Novak's lab had to review literature at a human pace to gather the information needed for his research. Now, AI is helping his team automate the process and review decades of ecological data quickly and efficiently.
During Oregon State University's 2026 Artificial Intelligence Week, researchers from across the College of Science gathered to demonstrate how artificial intelligence is transforming scientific discovery. From predicting the behavior of complex physical systems to uncovering patterns hidden in biological data, faculty shared how AI is helping answer questions that were once beyond the reach of traditional methods.
This story is the first of a two-part recap highlighting the presentations.
Douglas Walker, research associate in the Barbar Lab, studies intrinsically disorder proteins. These are a subset of proteins that, unlike traditional proteins, do not fold into a stable three-dimensional structure. Instead, they remain flexible and dynamic, enabling them to perform unique functions within cells. As a biophysicist, Walker is interested in learning how and why proteins assemble and interact the way they do.
To answer these questions, the Barbar lab often uses computational prediction methods to identify potential binding sites. However, traditional methods can often miss sites in disordered regions.
That was until the introduction of AlphaFold, a protein structure prediction algorithm that gained widespread adoption across the field in 2020. This tool allows researchers to predict multiple protein structures and interactions between multiple proteins. For Walker’s research, it opened new ways to investigate binding behavior that had previously been difficult to resolve experimentally.
“We all need to be AI scientists if we’re going to use AI methods. That’s not developing the models, but asking questions,” Walker said. “AI does not remove uncertainty, but it shifts it. The future isn’t scientists replaced by AI; it’s scientists who know how to think about AI.”
Marilyn Rampersad Mackiewicz, associate professor of chemistry, is integrating artificial intelligence with continuous flow systems to accelerate the design and production of nanomaterials for biomedical and environmental applications.
Her lab studies how engineered materials interact with living systems and how those interactions affect health and environmental outcomes. A key challenge is controlling how nanoparticles behave in biological environments, where properties like size, shape and surface chemistry can determine everything from toxicity to therapeutic effectiveness.
One of the biggest barriers is reproducibility. Traditional batch synthesis methods often produce heterogenous mixtures of nanoparticles, even when conditions like temperature, pH and stirring rates are tightly controlled. That variability makes it difficult to reliably study or scale materials for real-world applications.
To address this, her lab is building AI-enabled continuous flow systems that automate and accelerate materials discovery. One platform can generate liposome-coated nanoparticles in about 15 minutes, compared to roughly 24 hours using conventional methods, enabling rapid production and screening of hundreds of formulations.
A second custom-built system connects pumps, mixers and inline UV-visible spectroscopy to monitor reactions in real time as materials form. The setup produces continuous data streams that feed into analysis tools and machine learning models, allowing it to learn from each reaction and help guide subsequent experiments.
While AI supports experimental design and data analysis, Mackiewicz emphasized that scientists remain in control of the process. By combining automation with real-time analysis, the platform enables faster iteration across large libraries of nanomaterials. The system is already being applied to a range of projects, including materials for cancer therapies, stem cell imaging, glaucoma treatment and dental applications, with the broader aim of accelerating the path from materials design to clinical and practical use.
Marilyn Mackiewicz's autonomous platform integrates artificial intelligence, continuous-flow chemistry and real-time spectroscopy into a single workflow. This approach enables faster, more reproducible experiments while accelerating the discovery of new materials and chemical processes.
Mark Novak, associate professor of integrative biology, is using large language models and machine learning tools to help ecologists extract and predict biological data that would otherwise take years to compile manually.
Novak’s lab studies food webs and predator-prey interactions, focusing on understanding how energy and biomass move through ecosystems. Much of that work depends on information scattered across decades of ecological literature, including old diet surveys buried in historical journals and natural history publications.
One project focuses on predator stomach contents. While ecologists have long studied what predators eat, Novak’s research also examines what predators are not eating. Empty stomachs, he explained, can reveal clues about metabolic rates, mortality and ecosystem resilience, yet those observations are often left out of ecological databases.
Over the last decade, Novak’s lab manually assembled thousands of diet surveys from hundreds of papers. Now, working with undergraduate collaborators in the College of Engineering, he is using AI tools to automate much of that process.
The workflow combines machine learning classifiers with fine-tuned large language models trained on the lab’s existing dataset. The system first identifies papers likely to contain useful ecological data, then extracts relevant variables and context clues from the full text. Novak said large language models are especially valuable because key information can appear anywhere in a paper, not just in structured sections.
A second project uses machine learning to predict species’ body mass from names or taxonomy alone. The lab has compiled measurements for roughly 40,000 species and is using that data to estimate missing values for others.
The expanding datasets are already helping reveal long-term ecological patterns, including possible climate-related changes across decades.
Juan Vanegas, associate professor of biochemistry and biophysics, is using AlphaFold and related AI methods to explore how membrane proteins shift between different shapes and functional states.
His lab focuses on membrane proteins, which make up roughly 30% of all proteins and carry out essential roles in cells, including sensing environmental changes and supporting processes like vascular development and osmotic regulation. Many of these proteins are mechanosensitive, meaning they respond to physical force, but their structures are difficult to capture experimentally or computationally because they move between multiple conformations.
To study them, Vanegas has traditionally relied on molecular dynamics simulations, which use physics-based models to simulate atomic motion. While highly accurate, these simulations are extremely computationally expensive, sometimes taking days or weeks on supercomputers for a single protein system.
More recently, his lab has turned to data-driven approaches like AlphaFold, which was trained on large structural and genetic databases. Rather than generating just one predicted protein shape, Vanegas and his team are adapting the model to explore the different shapes a protein might naturally take. By making small changes to the information they provide the model, they can uncover a wide range of biologically realistic possibilities.
In some cases, these AI-generated conformations align with results from experimental structures and molecular dynamics simulations, suggesting the models are capturing physically meaningful behavior. Importantly, the approach is far less computationally intensive, producing results in minutes rather than days.
However, Vanegas noted limitations. The methods can behave like a black box, are difficult to control under specific environmental conditions and can be inconsistent depending on how closely a protein resembles training data.
Even so, he sees promise in using AI to better understand protein dynamics. The long-term goal is to move beyond simulating individual structures and instead using sequence information to predict the full range of shapes a protein can adopt, reducing or even replacing the need for costly simulations.
David Kikuchi, assistant professor of integrative biology, is deploying artificial intelligence to study how animals process information and make decisions in complex environments.
His lab focuses on animal behavior, particularly how organisms use visual information to navigate a constant stream of sensory input. Animals must filter out background noise, identify threats or food sources, maintain spatial awareness and remain vigilant for predators — all while integrating genetic, experiential and social sources of information.
Kikuchi is especially interested in how genetic information can shape “hardwired” behavioral responses, particularly in situations where learning from experience would be too risky. One example comes from studies of birds that show highly specific, innate responses to local venomous snakes they have co-evolved with. These birds can distinguish dangerous snakes from other similar-looking stimuli, suggesting they possess highly refined visual templates encoded in the brain.
To study these processes, his lab works with both wild and lab-raised birds, combining controlled behavioral experiments with field observations. Researchers measure movement, vocalizations and brain activity in response to simulated threats to understand how different types of information are integrated in real time.
AI tools are now central to this work. Kikuchi’s team is using machine learning and custom Python tools to analyze difficult behavioral data, including video of rapidly moving birds and thermal imaging that may reveal physiological signs of fear. Working with large language models, they are also developing code to extract patterns from complex datasets that would be difficult to process manually.
The lab uses pose estimation algorithms to track fine-scale movements of birds in real time, and acoustic analysis tools such as BirdNET and OpenSoundscape to classify and identify alarm calls. These systems convert sound into visual representations, allowing image-based machine learning methods to be applied to audio data.
In early stages of development, the lab is also applying machine learning tools to map whole-brain activity patterns, linking neural responses to behavior. Together, these approaches aim to connect genetic predisposition, sensory perception and neural activity to better understand how animals interpret and respond to threats in their environment.
Myriam Cotten, associate professor of biochemistry and biophysics, is exploring how artificial intelligence can accelerate the discovery and design of bioactive peptides, short chains of amino acids that can serve as therapeutics for a wide range of health applications.
Her lab focuses on peptides with antimicrobial, antiviral and antioxidant properties, particularly those that interact with cell membranes. Unlike many drug targets, bioactive peptides do not contain clear signatures or active sites that make them easy to identify or optimize, making discovery a slow and challenging process.
AI is beginning to change that. Machine learning tools can analyze the physical and chemical properties of known peptides and search genomes for new candidates with similar characteristics. Rather than relying solely on researchers to manually modify existing sequences, these approaches can identify promising peptide candidates and suggest ways to improve their activity.
Cotten is also using machine learning to analyze bioactive peptides derived from Pacific dulse, a red seaweed. Her lab breaks down a photosynthetic protein from the seaweed into complex peptide mixtures that show strong antioxidant activity. Working with collaborators in the College of Engineering, she is applying machine learning methods designed for small datasets to identify which peptide sequences are most likely responsible for those beneficial effects.
Recently, her lab worked on an antimicrobial peptide derived from fish that shows promise against pancreatic cancer. The team demonstrated that the peptide can target cancer cells while largely sparing healthy cells, and they are now seeking ways to reduce side effects by modifying its sequence.
Looking ahead, she sees opportunities for AI to assist in the design of improved peptide therapeutics by identifying which portions of a sequence are most important for biological activity and which can be altered to improve safety and effectiveness.
