Shaping healthcare’s future with genomic data
By engaging in what Hunter described as “relentless sub-classing,” his lab is able to create sub-classifications of individual molecules or mutations so that their descriptions do not contradict one another—even though various sources describing them might. Although conventional relational technologies are certainly employed for narrowly defined areas of genomic research, they are less effective in instances in which you “apply all the knowledge from outside to get a better understanding of what’s happening,” Plasterer said. “That’s where all of the reference databases that we’re getting from primarily public sources become so valuable.”
Artificial intelligence supports some of the basic processing required of computational biology and its genomic applications. Different facets of natural language processing (NLP), for example, are regularly deployed to glean value from text analytics necessary for parsing academic journals. “NLP, tagging and using that as a way to take unstructured and semi-structured sources of information to harmonize against the types of vocabularies that are needed for interoperability, that’s pretty common,” Plasterer said. “We’ll do that to take semi-structured pieces of information and make sure that everybody is calling a gene by the same name or variants by the same ID structure.”
Some of the more foundational gains in genetic research are directly attributed to AI. Hunter, who received a Ph.D. in AI from Yale in 1989 and founded a conference for AI application in molecular biology in 1993, said, “From the very beginning, analysis of genome sequences and gene sequencing in general, before we even got to genomes, has used AI technologies.” Other than NLP capabilities, the most common form of AI for genomic data is machine learning. Numerous machine learning techniques are deployed to expedite the process of sequencing genes. Machine learning is also a key capability for determining which pharmaceuticals are effective for patients suffering from specific conditions such as cancer.
According to Hunter, a new class of immunotherapy drugs has had varying ranges of effectivenesson different cancer types. “There’s been a lot of machine learning approaches to try and figure out which cancers are going to respond well to this immunotherapy, and those are inching their way to clinical use,” he said. The hidden Markov model, which is routinely used to identify points of similarity and differences for gene sequencing, was originally built for speech recognition purposes. “A lot of our most fundamental understanding of how biology works is based on these AI techniques,” Hunter said.
Analyzing genomic data can enable researchers and clinicians to personalize healthcare treatment. Perhaps the sole exception is the creation of therapies based on specific genetic mechanisms applicable to many conditions. Still, focusing on individual mechanisms is another dimension of tailoring treatment for specific patient groups. Those groups should spend less time awaiting treatment that works while reducing time spent on options that don’t.
Applications of genome sequencing can also impact preventative medication. According to Al-Siddiq, in the future genome sequencing could “identify risk factors for an individual or what they’re at risk for, and we can start much earlier to push the issue out so it doesn’t hit them when they’re 40 or 70. Or, we can fix it altogether or change their environment so they’re never afflicted with it.”