When the first published drafts of the human genome appeared 20 years ago, scientists proclaimed that an outpouring of genomically informed medical advances was imminent. For example, Francis S. Collins, MD, PhD, then director of the National Human Genome Research Institute, and Victor A. McKusick, MD, professor of medical genetics at Johns Hopkins University, predicted that by 2020, the impact of genetics on medicine would be “widespread.” They shared more detailed expectations in a paper titled, “Implications of the Human Genome Project for Medical Science” (JAMA 2001; 285(5): 540–544).
“The pharmacogenomics approach for predicting drug responsiveness will be standard practice for quite a number of disorders and drugs,” the scientists wrote. “New gene-based ‘designer drugs’ will be introduced to the market for diabetes mellitus, hypertension, mental illness, and many other conditions. … By 2020, it is likely that every tumor will have a precise molecular fingerprint determined, cataloging the genes that have gone awry, and therapy will be individually targeted to that fingerprint.”
For the most part, these expectations remain unrealized, mainly because the task of translating human genome insights into precision medicine practice has proven to be vastly more complex than scientists anticipated in 2001. Today, 20 years later, most scientists are reluctant to offer bold predictions. Instead, they ask questions, such as: What obstacles remain to the full realization of the promise of the Human Genome Project (HGP) for precision medicine? This very question was taken up by scientists who participated in a roundtable session at Genesis 2020, a virtual event held last December.
A shift in perspective
HGP-inspired expectations are hard to meet because few diseases are due to a single gene. More typically, a disease is connected to multiple gene variants. Also, many putative disease-linked variants fall into noncoding regions of the genome, raising more questions than answers.
“If you observe what the gene alteration is between people who have a particular disease, that doesn’t necessarily mean that you’ve identified the key driver for that disease,” said Mohammad Afshar, PhD, president and CEO of Ariana Pharma, a provider of clinical and biomarker data analytics services. “Genomics has been used initially in a more descriptive way, and what we need to do now is use it in a predictive way.”
In translational genomics, one of the biggest remaining obstacles is simply a lack of awareness of what tools and technologies are available for leveraging the power of the human genome. This is certainly the case in oncology, Afshar noted. “A number of clinical trials are still using very limited data because there isn’t the awareness that it’s relatively easy to get genomic data for patients,” he continued. “Also, there isn’t the awareness that you can analyze and integrate that kind of data to accelerate getting the drug approved for the right patients.”
That sentiment was echoed by Joanne M. Hackett, PhD, head of genomic and precision medicine at IQVIA. She suggested that people need to start thinking about precision medicine and genomic medicine “as a comprehensive picture as opposed to a magic bullet that’s going to solve the problem today.”
According to Hackett, the HGP has yielded major results already and continues to do so—it’s simply taking longer than expected. One example of a technology enabled by the HGP would be the drug combination encorafenib and cetuximab, which was approved in 2020 for colorectal cancer patients with the BRAF V600E mutation.
“That’s a really nice example of drug development, which in and of itself can often take many years, sometimes easily 15 to 20 years and millions and sometimes billions in funding to be able to make that happen,” Hackett elaborated. “You can really make those discoveries only when you have enough data amassed to be able to see those patterns.”
Still, said Hackett, advances in genomics are hampered by what she refers to as “plumbing.” That includes systems such as electronic medical records that can capture real-world information such as height, weight, smoking habits, and other data that can help sort out the effects of disease-related genes from the effects of lifestyle and environment.
Hackett added that much more data is still needed: “You can derive insights only if you have data from a colossal number of patients. No one really knows what that number is. But it’s not 500 patients. It’s not 5,000. It’s not even 5 million. We need as much as possible.” Ultimately, Hackett insisted, it will require public-private partnerships to build sufficiently large databases, as well as a process of building trust with people that their data will be safe and used responsibly.
Building diversity in genomic data
An IQVIA Institute report, “Understanding the Global Landscape of Genomic Initiatives,” notes that worldwide, 38 million genomes have been analyzed. That number, which is expected to reach 52 million by 2025, includes 1.5 million whole genomes from Europe and 780,000 from North America. Compilation efforts include the European 1+ Million Genomes initiative.
Current genomic data does not represent global populations well, particularly those in Asia, Africa, and South America. Although Asia makes up 60% of the global population, only 6 million Asian genomes are expected to be sequenced by 2025. An initiative aimed at expanding the diversity of available genomes has been undertaken by Global Gene.
The company, which hopes to facilitate the delivery of the next generation of therapeutics, is currently executing the largest gene sequencing program for populations from the Indian subcontinent, which is home to 4,500 different ethnicities, or almost 50% of the world’s ethnic diversity. To illustrate how important understanding this sort of diversity is, Sumit Jamuar, co-founder and CEO of Global Gene, said that some people in South Asia were found to not experience pain. Their insensitivity to pain, further research indicated, is due to a particular variant in the gene SCN9A. Targeting the gene with analgesics could lead to a totally new treatment paradigm for severe pain.
“These insights open up a new way of discovering and developing drugs,” explained Jamuar. The goal is to curate very high-fidelity disease cohorts combining genetics and phenotype, enriched with omics data to greatly accelerate artificial intelligence (AI)-driven therapeutics discovery and development.
“If you think of it as an engineering system, can we make disease optional?” asked Jamuar. “Suppose we have an insight in terms of our propensity for a particular disorder. If the disease has not yet manifested, we have an opportunity to intercept it before it does so. If the disease has manifested, we have an opportunity to personalize the treatment so that it’s more effective, which is where things like pharmacogenomics come into play. Or, to take this one step further, can we develop the personalized therapeutics to treat the disease based on genomics.”
The result, Jamuar asserted, will be a democratization of healthcare: “When you think about the way these technologies are being adopted, you see the potential to fundamentally transform healthcare into a completely new paradigm for billions of people.”
Using AI to mine genomic data
Another key to realizing the potential of the HGP is computing power, said Stephen Minger, PhD, director of SLM Blue Skies Innovations, a consulting firm that provides expert analysis in emerging healthcare technologies. He emphasized that the many layers of omics data—including genomics, transcriptomics, proteomics, metabolomics, and microbiomics data—require massive processing power to analyze.
Referencing efforts like the 1+ Million Genomes initiative, Minger noted, “This is really going to be a gold mine of information, assuming that we have synthetic intelligence to do all the heavy lifting.” He added that instead of systematically analyzing genomes, one may rely on synthetic or artificial intelligence systems. In his words, “You just turn them loose and let them data mine.”
To illustrate the usefulness of that kind of approach, Minger described how AI could help untangle the interaction between the gut microbiome and the efficacy of cancer drugs. “Microbiomic data is telling us that patients may or may not respond to given drugs simply because of flaws in their gut and how it’s metabolizing the drug in one person compared to another,” he said. “We couldn’t grow 99% of the bugs that are in our gut. And now, we don’t need to. We can just shotgun sequence the microbiome and pull out, based on homology, all of these different bugs.”
Then, using an unbiased machine learning approach, an AI system could find associations that would be difficult or impossible to locate using targeted, hypothesis-based experimentation. “The machines can plow through this a million times faster than humans can,” Minger elaborated. “If the machines can sort out stuff that’s too much data for us to dig through, you can then test those hypotheses.”
This kind of computing power wasn’t available when the final draft of the HGP was released in 2003, but it’s becoming available now. Minger referenced Deep Mind, an AI subsidiary of Alphabet (Google’s parent company). Deep Mind’s neural network–based AI has been put to work on some biotech-related projects. One of them is a joint research partnership with Moorfields Eye Hospital that showed AI could correctly analyze three-dimensional scans of the eye and recommend treatment with accuracy equal to that of highly trained ophthalmologists.
Increased computing power is also enabling progress in a discipline that’s been somewhat neglected in the genomic era due to its complexity: proteomics. In fact, computing power is being used to take studies of proteomics down to the single-cell level. One such study, the PIVOT-02 Phase I dose-escalation trial, was sponsored by Nektar Therapeutics and Bristol Myers Squibb. To evaluate 38 patients who had metastatic melanoma, the scientists who conducted the study made use of technology from IsoPlexis. They were able to predict progression-free survival based on single-cell proteomic readouts.
According to IsoPlexis’ CEO, Sean Mackay, the groundwork for such studies was laid by the HGP. He added that in the PIVOT-02 study, exploratory biomarker analyses of baseline tumor biopsies identified immune signatures that potentially enrich for response in patients with metastatic melanoma. Specifically, the biomarkers correlated with progression-free survival in patients treated with bempegaldesleukin plus nivolumab.
Detailed findings, Mackay noted, were presented at the Society for Immunotherapy of Cancer’s 2020 Annual Meeting and in the journal Cancer Discovery (Diab et al. 2020; 10(8): 1158–1173). “A key biomarker like that in a big clinical study,” he explained, “shows what you can do if you get single-cell proteomic resolution using our platform.” That type of information can then be pieced together with genomic data to characterize something more complex such as the immune response.
Although the readout of the HGP initially provided a wealth of information, and although the precision medicine impacts are being seen 15–20 years later, we can’t lose sight of the bigger picture, Mackay insisted. “We know that computing advancements that came during the 20th century evolved based on advances in physics research,” he continued. “Similarly, the research and data that emanated from the HGP is giving us a wealth of information to apply to precision medicine today, and translating that research just takes time.
“You see impacts for creating new therapeutic cells, creating new products in synthetic biology, and synthesizing reagents for better diagnostic tests. All of this is because of the genome project work that was done 20 years ago, and now, as we develop advanced medicines, we realize that we’re standing on the shoulders of giants.”