CMU Computational Biologists Harmonize Massive Genomic Datasets To Target Root Causes of Disease

Researchers in Carnegie Mellon University's Kenneth C. Griffin School of Computer Science are uniting different biological datasets into a single, shared system to help scientists develop targeted treatments for rare and complex diseases. Their work could lead to more targeted or rapid disease treatment in the future.

The human body contains hundreds of cell types and tens of thousands of genetic variations. Over the past two decades, sequencing technology has allowed scientists to identify roughly 50,000 to 70,000 genetic variants linked to human traits and diseases, like height or Alzheimer's disease. But to develop effective drug therapies, scientists must identify and understand what specific cell type a genetic mutation affects and how that mutation alters what the cell does.

Martin Zhang, an assistant professor in the Ray and Stephanie Lane Computational Biology Department, received a multiyear, $2.3 million grant from the National Institutes of Health (NIH) through its Supporting Talented Early Career Researchers in Genomics program to tackle this open problem.

"A disease-associated genetic mutation is usually defined within a one-megabase-long window of nucleotides called a locus," Zhang said. "Within this window, there are usually 30 to 50 genes. You want to make the right choice among those genes as to which one is the correct target. When developing a disease treatment, if you pick the wrong gene, changing that gene will not change anything about the disease outcome, wasting time and taxpayers' money."

It's not simply about understanding human DNA or messenger RNA (mRNA), which tells researchers which genes are actively expressed at a given moment. The sea of data Zhang wants to unify includes the physical shape of DNA in a cell, how neighboring regions impact genes, and what happens when certain genes in specific cells are activated or disabled. Combining this information can lead to more targeted disease therapies because multiple data sources point to the correct target. This means when a specific genetic mutation is identified to address a disease, researchers have multiple data sources to validate that conclusion.

Zhang said the NIH grant supports roughly 10 subprojects spanning three core directions: identifying the specific cell types where disease mutations occur; mapping those mutations directly to the genes they regulate without introducing bias; and integrating all these diverse biological data sources into a single, united framework. The researchers aim to develop a unified computational system that processes these different biological data sources all at once. By translating diverse datasets into a shared digital language, the platform allows researchers to evaluate the entire biological picture simultaneously rather than inspecting clues individually.

"The holy grail is to cure disease by looking at the mechanisms of genetic variants to see if we can intervene to alter the course of the disease," Zhang said. "For that, we need information beyond just the identity of the genetic mutations. The broader view of the biological question I am trying to answer is: Can I bring more insights into genetic associations so we can turn those insights into therapeutic treatment developments?"

This work can also tell researchers about human evolution, an area where Zhang's work is already offering that insight.

In a paper published in Nature Genetics, researchers detailed a new discovery called linkage masking. Traditionally, scientists assumed genetic mutations act independently regarding natural selection. But the research team, which included Zhang, discovered that genetic mutations emerge in pairs. For traits like height, being extremely tall or extremely short can be a disadvantage. If a person carries both a height-increasing mutation and a height-decreasing mutation, they cancel each other out, allowing both mutations to persist together in the population.

"Practically, it tells us that when we identify a causal mutation for a disease, we should expect another mutation nearby acting in tandem," Zhang said. "They often appear in pairs or groups, rather than as lone wolves."

Zhang said this work relies on the wealth of information that already exists at researchers' fingertips, from DNA sequencing to spatial maps of single cells.

"I like to think of us as detectives," he said. "Gathering and harmonizing those clues is the core effort and we follow them until all arrows point to the exact same target."