Processing power

Ferhat Ay, Ph.D., is an Associate Professor at La Jolla Institute for Immunology (LJI). He isn’t a wet-lab scientist. He doesn’t dissect tissue or peer into microscopes. 

And yet, Dr. Ay sees immune cells up close every day. 

As a data scientist, Dr. Ay develops algorithms and bioinformatics tools at LJI to uncover the hidden lives of immune cells. His laboratory uses these tools to shed light on how different regions of the vast human genome contribute to cancer, autoimmune disease, and more.

It’s a daunting task. “We work with single-cell datasets, where one cell at a time is measured for all sorts of molecular activity,” says Dr. Ay. “These datasets are in the order of tens of millions, 100 million, data points or so.”

No human could ever sort through such vast data. So, Dr. Ay became a leading expert in the use of artificial intelligence (AI) to understand the intricate workings of the human genome. 

This isn’t your everyday AI. Dr. Ay has seen the bizarre answers generated by ChatGPT and other popular AI platforms. Mistakes like those are not an option when you study real diseases that affect real people. 

“We are working with tight margins,” says Dr. Ay. “We don’t have that kind of error tolerance.” 

Immunologists need incredibly precise AI tools. LJI scientists are leading the way in building and testing those tools for a new era of research.

How to teach a machine

LJI Professor Bjoern Peters, Ph.D., has been part of the AI revolution since the very beginning. His team has spent the last 20 years developing sophisticated AI tools called machine learning models. 

Machine learning models “learn” from scientific datasets. They sort through data to find patterns and trends, then try to complete the pattern if any data are missing. “Machine learning is about learning from examples,” says Dr. Peters. 

The more data you input, the smarter the model gets. A good machine learning model can make startlingly accurate predictions and help answer complicated scientific questions. 

Dr. Peters is currently using a machine learning model to study bacterial pneumonia. His model harnesses existing data to learn how T cells recognize protein markers, called epitopes, on Streptococcus pneumoniae bacteria. The model then looks for patterns to predict where scientists might find bacterial epitopes that they haven’t studied yet. 

“This bacterium makes around 2,000 proteins,” says Dr. Peters. “With epitope prediction, we can look for the proteins that are most easily recognized by the immune system.” Once Dr. Peters has those predictions, he can collaborate with other scientists to measure actual T cell responses in a lab setting. 

This kind of AI approach makes it possible for scientists to narrow the scope of a big project, saving them valuable time and funding. 

Of course, AI tools are only as good as the data they learn from. Quality control is key. Dr. Peters works closely with curators who comb through scientific papers to find epitope data related to viruses, bacteria, and allergens. The team makes sure every data point is accurate and well defined before it ever meets an AI tool. Dr. Peters and his colleagues offer these data freely through the Immune Epitope Database (IEDB). In fact, the IEDB is only one of several LJI-hosted databases powering new AI approaches to vaccine research and much more.

New discoveries fuel vaccine research

LJI Research Assistant Professor Alba Grifoni, Ph.D., is using AI to study viruses with pandemic potential. Her goal is to help guide the development of “universal vaccines” that protect against entire viral families. 

Dr. Grifoni investigates what a family of viruses has in common, then helps develop vaccines that boost T cell responses that fight those common targets. 

In a recent paper, Dr. Grifoni extracted IEDB epitope data to compare how human T cells recognize different coronaviruses. If T cells can recognize common cold coronaviruses, then they should be able to recognize SARS-CoV-2 (the coronavirus that causes COVID-19). After all, the two viruses ought to share some family resemblance. 

Dr. Grifoni used AI in a different way for this study. Rather than make predictions, Dr. Grifoni used AI to weed out irrelevant datapoints. She didn’t want the massive SARS-CoV-2 dataset (big outbreaks equals big datasets) to throw off the whole analysis. 

Dr. Grifoni’s study revealed a hidden set of similarities between SARS-CoV-2 and other coronaviruses. She hopes future vaccines will be able to boost T cell responses against these shared targets to give us significant “pan-coronavirus” immunity. 

“The idea is that if a new coronavirus emerges, we may not be able to protect people from infection, but we might be able to protect them from hospitalization,” says Dr. Grifoni.

AI + cell images

LJI Assistant Professor Miguel Reina-Campos, Ph.D., uses AI tools to investigate how specialized T cells defend our tissues from cancers and other threats. 

For this research, Dr. Reina-Campos needs to analyze microscopy images from tissue samples. Many, many tissue samples. “Our task is to find patterns in those images,” says Dr. Reina-Campos. “So we use an AI software that leverages a trained dataset—a trained model—to find these patterns.” 

Once the AI software identifies an unusual pattern, Dr. Reina- Campos and his team can zoom in to study that bit of tissue in more detail themselves. Dr. Reina-Campos can even use AI to add more layers of information. “We can use AI to classify cell types, decode their language, and understand how they interact with one another,” he says. “Are the cells fighting, competing, or cooperating?” 

Establishing cellular connectivity networks enables Dr. Reina-Campos and his team to see the bigger picture and understand how immune cells work together. Dr. Reina-Campos is currently working with a global team called the Immunological Genome Project (ImmGen) to use AI tools to map the daily lives of all immune cells in mouse tissues. This project, called ImmGenMaps, is key to understanding how our immune cells jump into action to fight tumors and other threats.

AI in the doctor’s office

LJI scientists are using AI to make breakthroughs in personalized medicine, too. 

LJI Bodman Family Assistant Professor Tal Einav, Ph.D., develops algorithms that give AI tools their predictive power. His work may lead to real advances in the doctor’s office.

In a recent study, Dr. Einav put a new machine learning model to work to pinpoint key differences between people who are “strong responders” to annual flu vaccines and people who are “weak responders.” 

Dr. Einav discovered that the best way to predict how a person will respond to an upcoming flu vaccine is to measure their antibody responses to the flu vaccine strain used the year before, and the year before that, and the year before that. 

Dr. Einav thinks it may be possible to develop blood tests so clinicians can identify which patients are strong flu vaccine responders and which are weak vaccine responders. Doctors could then offer their patients different vaccine options based on their responder type. “The hope is that we could take a drop of your blood and say, right now, ‘Which are you going to be?’” Dr. Einav explains. “That’s the dream.” 

Dr. Ay imagines a future where these kinds of AI tools are part of a doctor’s toolbox. As he uses AI to explore the genome, he sees doctors harnessing similar tools to help their patients. 

“Let’s say you’re a doctor working with a kid who has a rare disease, or a mutation of unknown significance,” says Dr. Ay. “If we can develop these types of AI models, we can enable clinician-scientists to put data in context in a way they never could before.”