AI and Genomics Reshape Pulmonary Biology Research

Jichao Chen, PhD, the new division director of Pulmonary Biology at Cincinnati Children’s, is leading a push to infuse artificial intelligence (AI) and computational genomics into research.

By harnessing technology, Chen says researchers can use AI and computational models to run complex, multidimensional calculations and predictions that the human brain cannot complete on its own.

“AI can handle the extra dimensions and algorithms by adding software and memory,” Chen says. “It doesn’t qualitatively change the nature of its work.”

Understanding Biology’s Multidimensional Multiverse

“The complexity in biomedicine is beyond human intelligence for a few reasons,” he says.

First, this is a multidimensional problem. The simplest way to frame it is that there are 20,000 genes in the human genome. Each gene represents a dimension, as now measured with single-cell omics technology. Humans struggle to wrap their minds around dimensions beyond space and time.

Secondly, the genome works within a multiverse—much like the Marvel movies, Chen says. In biology, the characters are the genes. And each cell is its own universe.

“Those genes are all doing their own thing in their own way in diverse cell types, but they all exist in harmony within the whole body,” Chen says.

This multidimensional multiverse nature of biology is why so many experiments fail, Chen says, because scientists can’t accurately predict the results without computational help.

Merging Human and Machine Understanding

Enter the capabilities of AI. Researchers can use natural language (our spoken language) and what Chen calls nature language (biology’s language) to train AI to make predictions.

On the clinical side, doctors’ notes and patient testing results can be used to train AI models to determine which patients in a NICU are most likely to have a genetic condition. Such results can be used to prescribe genetic testing for patients.

On the research side, AI models that treat a gene like a word and a cell like a sentence can be trained to make needed calculations.

“Just as you can force AI to learn certain words and predict which words should fill in a blank space, like with ChatGPT, you can do the same with genes,” Chen says.

This forced learning teaches an AI program which genes work together and how they’re regulated and function within a cell. Once an AI program learns the nature language from millions of normal cells the way ChatGPT learns from the internet, scientists can ask the program to predict what will happen when specific genes or groups of genes are removed.

Today, as such omics training data becomes increasingly available, AI programs are making rudimentary predictions about the consequence of gene editing.

Going forward, Chen sees “an opportunity to merge the machine understanding and the human understanding. Can we infuse decades of biomedical literature into machine knowledge and come up with a more intelligent understanding of biology?”

To Chen, this is the key to solving the problem of why many experiments and trials fail.

LungMAP Project Uses AI to Uncover Hidden Pathways of Lung Disease

AI and computational genomics also play a growing role in Cincinnati Children’s work on the multicenter LungMAP project.

Funded by the National Heart, Lung, and Blood Institute of the National Institutes of Health, LungMAP is an open-access reference resource of a 3D molecular atlas of both normal and diseased human lungs. The project makes data and reagents available to researchers.

Chen leads Cincinnati Children’s LungMAP work. It is one of six centers collaborating on the project, also serving as the data coordinating center that synthesizes and distributes information.

A chatbot uses AI to interact with researchers. Scientists can ask a question in natural language about a gene’s activity within a certain disease and get an “in silico,” data-based answer.

For example, the LungMAP team can predict what happens to lung tissue when a specific gene is deleted. Researchers can validate that response by looking at tissue samples to see the real-life biological changes.

Going forward, Chen hopes AI can chart the hidden pathways of disease that occur before it presents at a late stage.

“We can’t follow the same tissue over time, but computationally, we can reconstruct it,” Chen says. “If we can find biomarkers that might indicate a turn down the wrong path, we can create early interventions.”

(Published December 2024)

Read More from the Perinatal Institute