Mike Klein, CEO of the genomic health company Genomenon, told GEN Edge that the company’s tracking system identifies about 15,000 newly published, peer-reviewed research articles per week with genomics data—or roughly 750,000 per year!
But Klein said that one of the things that researchers, whether in industry or academia, have not been able to leverage very well is the millions of dollars in data that is captured in genetics articles because scientific publications are not well organized. While people are able to create a lot of data by sequencing a lot of patients, the question Klein is focused on is: what does this data mean?
As a data curator, Genomenon is trying to keep pace with the constant advances in genetics and genomics to be in a position to serve the drug discovery market and help diagnose and treat patients with rare genetic diseases and cancer.
To do so, Genomenon recently acquired Boston Genetics, a genomics interpretation and curation company, in late June 2023 as part of its plan to curate the entire human genome. The goal is to combine Genomenon’s AI-powered platform for genomic knowledge and tech-enabled scientific expertise with Boston Genetics’ team of genetic scientists to take the lead position and be the first company to curate the entire human genome.
Born out of need
Mark Kiel, MD, PhD, was working at the University of Michigan when he became frustrated with the slow pace and a large time commitment of manual genome curation, prompting him to found Genomenon, whose name means “born out of need” in ancient Greek. The manual effort required to search and analyze the rapidly expanding body of medical literature for disease-related genes and variants made next-gen sequencing (NGS) data interpretation inefficient and prone to error.
Kiel launched Genomenon to meet the growing demand for accurate and timely variant interpretation by combining human curation with novel machine-learning approaches that can keep up with the rapid pace of publication and surface the most pertinent data to researchers.
In 2017, Genomenon released its first product—the Mastermind Genomic Search Engine—to aid researchers in their quest to search across multiple diseases, genes, and variants in order to fill in the gaps in genomic evidence.
According to CEO Klein, they developed Mastermind after discovering that most businesses were using natural language processing repeatedly on research articles without any success. As a result, they developed their own language processor using a technique they coined “genomic language processing” (GLP).
According to Klein, “We frequently find that authors can describe a single variant in 200 different ways. Genes can be described using legacy nomenclature in dozens of different ways. So, being able to build a bespoke set of algorithms that could go through and recognize the hundreds of ways that all of these criteria are used to describe variants and genes can describe variants, and then making sure that we’re actually capturing disease-variant relationships in the right way, we had to add some biological logic to our genomic language processing to actually make sure that we were coming up with solid results.”
The GLP has the underlying indexing capability that gives Genomenon the ability to find these genetic associations of disease, pull that information out, and normalize all that. When accessing Mastermind, there’s the capability to type in dozens of different ways to describe a variant.
But Klein said that in 2019, pharma companies were approaching them because they didn’t want to click through individual gene variants—they really wanted to understand the entire genetic landscape of a specific disease. In other words, pharma companies wanted to provide data sets and have Genomenon curate these data sets for them.
“They wanted us to tell them every variant and every gene associated with a specific disease and give us an indication of the pathogenicity by clinical guidelines,” Klein said of the pharma companies. “Give us the functional consequences: is it a gain of function or a loss of function? What are the protein domains that these variants are driving?”
At that point, Genomenon began developing curated genomic landscapes. Over time, Klein started recognizing that this was a powerful solution, especially when it comes to stratifying clinical trials.
“If you think about finding the most likely responders, well, if you know the genetic drivers of the disease, you have the opportunity to increase the probability of success by selecting the right patients with the right genomic biomarkers and then leveraging that same data in the companion diagnostic (CDx) world for CDx development,” said Klein.
“As this market shifts from the platform side to really an information market, what the real world really needs is a curator.”
Genomenon has leveraged AI in developing its curation and indexing capabilities as well as in building a search engine. In the process, Klein realized that Genomenon had the opportunity to curate the “entire” human genome. By that, Klein means that he wants to identify every variant in every gene for every disease and the pathogenicity and the functional consequences of every one of those variants. He wants to put that information at the fingertips of pharma researchers to go beyond not just the clinical trial stratification and companion diagnostic development but actually reach into the drug discovery, be able to provide insight into the gene-disease relationship, and actually down to the variant level that they can use as they’re looking to really understand the genetic drivers of disease.
Klein believes that the NGS market is moving from what he calls a “platform play” into a “content play.”
“Netflix talked about their network reach and delivery vehicles, but when you talk about streaming today, we talk about content, content, content,” said Klein. “In the NGS market, you’re seeing a number of platform boxes coming in. Illumina is seeing pressure in the market from MGI, PacBio, Element Biosciences, and Ultima, who are driving the cost of sequencing and reagents down. As a result, the cost of turning whole genome sequencing—the actual chemistry—into data has dropped dramatically, and we’re seeing a lot more patients getting whole genome sequencing.”
To date, Genomenon has indexed the text of every identifiable genomics article. Klein said that there are about 9.2 million published, peer-reviewed research articles and 3 million supplemental data sets that have genomic data in them, and Genomenon has been able to pull out over 22 million genomic biomarkers or variants and their associations with diseases, phenotypes, and therapies.
Klein said that Mastermind has 100 times more genomic evidence and 50 times more variants than any other resource.
According to Klein, Genomenon has come to the realization that they cannot rely on AI alone to curate the entire human genome and keep it up-to-date. “AI can serve up some bad information and have a high false-positive rate, so we need to have the expertise on the backend,” said Klein.
Human genome curation needed a human element. In 2019, Genomenon started bringing in some of Boston Genetics’ team to start doing some genetic curation and ended up being their largest customer by the end of 2022. Now, the two have come together as a single entity.
“They approached us and asked what we thought about putting the two companies together,” said Klein. “It made perfect sense for us to couple the AI with genetics and genomics experts to lay out the curation of the entire genome. It’ll have the people and the technology to accomplish that mission.”
Klein hopes that Genomenon’s larger team and increased knowledge of genomics will immediately allow it to better assist its clinical, diagnostic, and pharmaceutical clients in their efforts to better understand the genomic drivers of genetic disease and cancer. The acquisition has expanded Genomenon’s genetic science team fivefold.
Today, Genomenon’s workforce consists of software developers, bioinformaticians, and genetic scientists. The acquisition also gives Genomenon the ability to offer affordable variant curation team extensions to genetic testing labs, which should reduce turnaround times.
“Boston Genetics does interpretation for some of the largest clinical labs across the United States,” said Klein. “We’re continuing that business—we’re not throwing that away. In fact, we have the capability to accelerate the interpretation services by leveraging some of the AI technology that we developed. We’re going to continue adding an additional revenue stream to the company, but we look to leverage that and expand that.”
So, this acquisition not only puts Genomenon ahead in the race to curate the entire human genome, but it also brings unprecedented actionable genomic insights for drug development programs as well as clinical diagnostic and newborn sequencing programs.
While all of this has been happening, Klein said that they’ve been incorporating machine learning and large language models (LLMs). That doesn’t mean they’re throwing ChatGPT at the data to provide answers; what Genomenon is doing is using the LLMs to sort the information in the right way so that their experts can get to the right information very quickly to make the right decisions to curate the data and be able to fill out their database and provide that to their users.
“We’ve developed the technology and the AI platform, but technology alone doesn’t get you to the true answers the researchers and clinicians need; it needs to be coupled with scientific expertise,” said Klein. “What this acquisition has provided us is another 75 people—scientific experts and genetic scientists—that we can bring into the company immediately. That was the real value added. It really enables our mission of curating the genome and puts us in a really strong position to be the first company to curate the entire genome.”