New Machine Learning Method Predicts Damaging Missense Variants

The premise of genomic medicine is that a person’s genomic characterization can be used to improve medical diagnosis, prognosis, and treatment. Each person, however, has millions of genetic variants, the vast majority of which have negligible impact on their health. How to determine which variants are relevant to a particular condition is a central issue in genomic medicine.

The issue is most pressing in the case of missense variants, which alter a single amino acid in proteins. Only about 20–30 percent of these mutations have a functional impact. Thus the question of how likely a variant is to change protein function—contributing to a health condition—is extremely uncertain for missense variants. As a result, most missense variants in clinical genetic testing are classified as VUS (variant of uncertain significance).

Yufeng Shen, PhD, an associate professor in the Department of Systems Biology and the Department of Biomedical Informatics, and his group have developed a new method for predicting which missense variants are potentially damaging. The method, called gMVP (graphical model for predicting Missense Variant Pathogenicity), uses one of the latest machine learning techniques, a graph attention model, to capture information relevant to predicting which variants are potentially damaging. Their paper, “Predicting Functional Effect of Missense Variants Using Graph Attention Neural Networks,” was published in Nature Machine Intelligence on November 15th, 2022.

Download PDF

Read or download the latest issue of the DSB annual newsletter. 

The seemingly chaotic bacterial soup of the gut microbiome is more organized than it first appears and follows some of the same ecological laws that apply to birds, fish, tropical rainforests, and even complex economic and financial markets, according to a new paper in Nature Microbiology by researchers led by Dennis Vitkup , PhD, associate professor of systems biology , at Columbia Univesrity Irving Medical Center .

One of the main challenges facing researchers who study the gut microbiome is its sheer size and amazing organizational complexity. Many trillions of bacteria, representing thousands of different species, live in the human intestinal tract, interacting with each other and the environment in countless and constantly changing ways.

"Up to now, it has been an open question whether there are any natural laws describing dynamics of these complex bacterial communities.”-Dr. Vitkup

The study’s discovery of multiple principles of gut bacterial dynamics should help researchers to understand what makes a gut microbiome healthy, how it may become perturbed in disease and unhealthy diets, and also suggest ways we could alter microbiomes to improve health. Read the full article in the CUIMC Newsroom. 

The study is titled “Macroecological dynamics of gut microbiota.” The other contributors are Brian W. Ji (Columbia), Ravi U. Sheth (Columbia), Purushottam D. Dixit (Columbia and University of Florida, Gainesville, FL), and Konstantine Tchourine (Columbia).

Congenital diaphragmatic hernia (CDH) is a severe birth defect. For babies born with CDH, their diaphragms are not developed properly, with some or all parts of the abdominal organs pushed into the chest. The displacement of these critical organs can have a significant impact on how the lungs develop and grow. 

Yufeng Shen, PhD, associate professor of systems biology

A recent study , led by Yufeng Shen , PhD, and Wendy Chung , MD, PhD, and their labs at Columbia University Irving Medical Center , investigated the genetic risk factors linked to CDH and analyzed data from whole genome sequencing and exome sequencing to determine novel mutations. The study also uncovered the link between CDH and additional developmental disorders. 

“Many babies with this birth defect also have lung hypoplasia or pulmonary hypertension and babies have difficulty breathing. Even with advanced care available, the mortality rate is still about 20 percent,” says Dr. Shen, associate professor of systems biology at Columbia, with a joint appointment in the Department of Biomedical Informatics . 

“One hypothesis is that the lung condition is not necessarily caused by the physical compression on the developing lungs in the chest,” explains Dr. Shen, “it can be caused by the same genetic defect that causes CDH. Finding those genes is absolutely necessary to improve care and develop effective treatment in the long run.”   

Scientists have been aiming to identify new risk genes in CDH—and other developmental disorders—with  the hope that with improved genetic diagnosis more tailored or long-term care for patients born with this defect could be provided, as well as potential targets for intervention down the road. 

The illustration above depicts with an example of four genes, how knowing how variable genes are in the normal population helps to find candidate disease genes in a patient. Above, top: Tuuli Lappalainen, PhD; bottom: Pejman Mohammadi, PhD.

For individuals with rare diseases, getting a diagnosis is often a long and complicated odyssey. Over the past few years, this has been greatly improved by genome sequencing that can pinpoint the mutation that breaks a gene and leads to a severe disease. However, this approach is still unsuccessful in the majority of patients, largely because of our inability to read the genome to identify all mutations that disrupt gene function.

In a new study published on October 10 in Science , researchers from New York Genome Center , Columbia University , and Scripps Research Institute propose a solution to this problem. Building a new computational method for analyzing genomes together with transcriptome data from RNA-sequencing, they can now identify genes where genetic variants disrupt gene expression in patients and improve the diagnosis of rare genetic disease.

The new method introduced in this study, Analysis of Expression Variation or ANEVA, first takes allele-specific expression data from a large reference sample of healthy individuals to understand how much genetic regulatory variation each gene harbors in the normal population. Then, using the ANEVA Dosage Outlier Test, researchers can analyze the transcriptome of any individual – such as a patient – to find a handful of genes where he or she carries a genetic variant with an unusually large effect compared to what healthy individuals have. By applying this test to a cohort of muscle dystrophy and myopathy patients, the researchers demonstrated  the performance of their method and diagnosed additional patients where previous methods of genome and RNA analysis had failed to find the broken genes.

Time lapse of a developing drosophila embryo. (Credit: Carlos Sanchez-Higueras/Hombría lab/CABD)

Every animal, from an ant to a human, contains in their genome pieces of DNA called Hox genes. Architects of the body, these genes are keepers of the body’s blueprints; they dictate how embryos grow into adults, including where a developing animal puts its head, legs and other body parts. Scientists have long searched for ways to decipher how Hox genes create this body map; a key to decoding how we build our bodies.

Now an international group of researchers from Columbia University and the Spanish National Research Council (CSIC) based at the Universidad Pablo de Olavide in Seville, Spain have found one such key: a method that can systematically identify the role each Hox gene plays in a developing fruit fly. Their results, reported recently in Nature Communications , offer a new path forward for researchers hoping to make sense of a process that is equal parts chaotic and precise, and that is critical to understanding not only growth and development but also aging and disease.

“The genome, which contains thousands of genes and millions of letters of DNA, is the most complicated code ever written,” said study co-senior author Richard Mann , PhD, principal investigator at Columbia’s Mortimer B. Zuckerman Mind Brain Behavior Institute and a faculty member in the Department of Systems Biology . “Deciphering this code has proven so difficult because evolution wrote it in fits and starts over hundreds of millions of years. Today’s study offers a key to cracking that code, bringing us closer than ever to understanding how Hox genes build a healthy body, or how this process gets disrupted in disease.”

Read the full article at the Zuckerman Institute. 

Protein engineering is a relatively young field that creates new proteins never seen before in nature. Today’s protein engineers usually create synthetic proteins by making small changes to the gene that encodes a naturally occurring protein. The variety of synthetic proteins range from stain-removing enzymes that have improved detergents to a long-acting insulin that’s used by millions of people with diabetes.

But two big unsolved challenges for protein engineers remain: The gene encoding the synthetic protein needs to be contained to prevent escape into other organisms and the gene needs to resist mutating over time so the protein doesn’t lose its function.

By merging two genes into a single DNA sequence, Columbia University synthetic biologists have created a method that could prevent human-engineered proteins from spreading into the wild, as well as stabilize synthetic proteins so they don’t change over time. The work, recently published in Science, was developed by Harris Wang, PhD, assistant professor of systems biology, with graduate student, Tomasz Blazejewski and postdoctoral scientist, Hsing-I Ho, PhD. 

In devising the method, the researchers were inspired by overlapping genes in viruses. When two different genes overlap, they occupy the same sequence of DNA. But the genes are read in different frames so that two different proteins are produced.

In overlapping genes, a random mutation in the sequence may not affect one gene, but it’s likely that it will harm the second gene.

“Overlapping genes essentially lock in a specific DNA sequence, and we thought we could exploit this idea for synthetic biology ...Ten years ago, we didn’t have the technology that would make this possible,” says Dr. Wang. “We didn’t have enough sequences in the database to make informed predictions and we didn’t have a way to synthesize long DNA sequences for testing our predictions.”

Chaolin Zhang, PhD, associate professor of systems biology

A new study by researchers in Dr. Chaolin Zhang’s lab at Columbia’s Department of Systems Biology details a novel computational method that models how RNA-binding proteins (RBPs) recognize specific sites in the target RNA transcripts, precisely and accurately. The researchers’ findings include identification of entirely new motifs (RNA sequence patterns), and their research in complex RNA regulation contributes to our understanding of the molecular basis of disease and conditions, and down the road, could aid in the development of targeted therapies. 

The study, led by Dr. Zhang, associate professor of systems biology, with senior co-authors Suying Bao, PhD, and Huijuan Feng, PhD, appears today in Molecular Cell. 

RNA has traditionally been considered mere “messengers” that transfer genetic information from DNA to proteins that ultimately carry out cellular functions. However, it is now increasingly appreciated that RNA can be tightly regulated to control gene expression and diversity protein products. RNA-binding proteins (RBPs) are at the center of such regulation, with important roles in many cellular processes, including cell function, transport, and location. Gaining mechanistic insights of the binding specificity of RBPs in a genome-wide scale helps advance our knowledge of gene regulation.

“RNA-binding proteins are crucial for gene expression,” says Dr. Feng, coauthor of the study and post-doctoral research scientist in the Zhang lab. “RNA is heavily regulated, and when this regulation goes wrong, instabilities or disease could occur.”  

One of the structural variations detected in Anaerostipes hadrus, which is deleted in ~40% of the population (top), and associated with higher disease risk. Genes in this region (bottom) code a composite inositol catabolism - butyrate production pathway, potentially supplying the microbe with additional energy while supplying the host with butyrate, previously shown to have positive metabolic and anti-inflammatory effects. (Credit: Korem lab)

Our gut microbiome has been linked to everything from obesity and diabetes to heart disease and even neurological disorders and cancer. In recent years, researchers have been sorting through the multiple bacterial species that populate the microbiome, asking which of them can be implicated in specific disorders. But a paper recently published in Nature addressed a new question: "What if the same microbe is different in different people?" The study was co-led by Dr. Tal Korem , assistant professor of systems biology and core faculty member in the Program for Mathematical Genomics at Columbia University Irving Medical Center . 

It has been long known that the genomes of microbes are not fixed from birth, as ours are. They are able to lose some of their genes, exchange genes with other microorganisms, or gain new ones from their environment. Thus, a detailed comparison of the genomes of seemingly identical bacteria will reveal sequences of DNA that occur in one genome and not others, or possibly sequences that appear just once in one and several times over in others. These differences are called structural variants. Structural variants - even tiny ones - can translate into huge differences in the ways that microbes interact with their human hosts. A variant might be the difference between a benign presence and a pathogenic one, or it could give bacteria resistance to antibiotics.