Latest News

Next Generation Sequencing, Precision Oncology, & Will AI Take Over The Lab?

By Leah Rosenbaum

February 22, 2018 | Next-generation sequencing, precision medicine, AI doctors and…pigeons? These were just some of the many topics covered in the Molecular Medicine track of the 25th annual Molecular Medicine Tri Conference*. Molecular medicine and diagnostics experts from around the world came to take part in the conference, which spanned three sunny days in San Francisco. Topics covered included Digital Pathology, Molecular Diagnostics, and Molecular Diagnostics for Infectious Disease, among others.

Next Generation Sequencing (NGS) was a big topic at the beginning of the week, spanning several different conference tracks. Although the Sanger sequencing technique, developed by Frederick Sanger and his team in 1977, has still remained popular, NGS is the new frontier for high-throughput DNA sequencing.

Monday morning in the Molecular Diagnostics track began with a discussion about NGS technology and its use in the clinical setting. Maria E. Arcila, pathologist and Director of the Diagnostic Molecular Pathology Laboratory at Sloan Kettering, spoke about how NGS is moving from a research to a clinical setting—or, as she said, “from bench to bedside.” This move means new challenges for NGS technology, and new guidelines and regulations to help it be effective in clinical settings. Arcila spoke about a paper (DOI: https://doi.org/10.1016/j.jmoldx.2017.01.011) she and her colleagues published last year in the Journal for Molecular Diagnostics, detailing guidelines for NGS sequencing in oncology testing.

Acrila was followed by talks from several physicians at the University of Pittsburgh, including Marina N. Nikiforova, Director of the Molecular and Genomic Pathology Laboratory at the University of Pittsburg Medical Center. NGS sequencing, Nikiforova said, “is especially important to understand for cancer cells.”

Nikiforova gave the example of a 43-year old woman with headaches and seizure-like activity who was found to have a tumor in her brain. The tumor was biopsied and sent for analysis of gene mutations. When the analysis returned, physicians saw mutations in several genes. The site of these mutations allowed them to make more informed decisions when diagnosing and treating the patient, and gave them a more accurate estimation of her prognosis. In other words, NGS technology can turn oncology into ultra-precision medicine.

Monday afternoon the NGS talks continued at the Molecular Diagnostics for Infectious Disease tracks. Duncan MacCannell, Chief Science Officer for the CDC’s Office of Advanced Molecular Detection, discussed how NGS sequencing is used in a public health setting. Since the creation of the Advanced Molecular Detection (AMD) and Response to Infectious Disease Outbreaks program in 2014, MacCannell says that he and his team have been able to use NGS technology to respond to public health threats faster and with more precision.

One example he gave was with foodborne disease surveillance. Soon after the AMD began, they started a nationwide listeriosis surveillance program. Listeriosis, a serious infection caused when food is contaminated by the bacterium Listeria monocytogenes, affects about 1,600 Americans each year and kills 260. It is especially dangerous for pregnant women. It’s a disease with potentially “devastating” consequences, said MacCannell, but now the CDC is able to sequence all of the cases that occur each year. This allows them to link listeriosis outbreaks to specific food sources, which enables companies to recall potentially contaminated products before they make more people sick.

 

AI Steals The Show

One of the most packed Molecular Medicine talks of the first day was the late-afternoon talk by Andrew Beck, co-founder and CEO of PathAI, on how artificial intelligence can aid in precision pathology. It would be a topic that would come up again and again over the course of the next two days of the conference, and there were many different opinions on the subject. Beck began his talk by discussing the challenges of pathology, chiefly that different pathologists can interpret test results in different ways. “In some of the most difficult areas,” Beck said, “it’s like a coin flip.”

With PathAI, Beck aims to create Deep Learning (DL) algorithms that can help pathologists do their job with more precision and accuracy. Beck cited a study published in JAMA this past December (DOI: 10.1001/jama.2017.14585.), which assessed DL algorithms and pathologists side by side to see which group made more accurate diagnostics with breast cancer slides. While pathologists had an error rate that ranged between 3.5-42%, the AI had an error rate of only 0.65%. PathAI and other AI companies, Beck said, could help pathologists by pre-screening slides and highlighting problem areas for pathologists to look at more closely. (For a similar perspective, see a Definiens commentary from last May)

The Q & A discussion after Beck’s talk was lively, and highlighted factions within the pathologist community. While one pathologist agreed with Beck and said that an AI would make it easier for her lab to agree on a diagnosis, another pathologist felt strongly that Beck exaggerated the problems in traditional pathology.

The debate over AI continued Tuesday and Wednesday. On Tuesday morning in the Digital Pathology track, Nigel Lee, Chief Algorithmic Officer at Corista, and Jason Baron, Assistant in Pathology at Massachusetts General Hospital and Assistant Professor of Pathology at Harvard Medical School, continued to tout the benefits of integrating AI with pathology. “I think we’ve really reached this point where the rate that we produce data is exceeding our capacity as clinicians, pathologists, and technologists,” Baron said. This is where AI can come in to help—to filter through data, and enable better and more precise decision making.

One speaker, however, didn’t seem convinced that AI would soon be able to replace humans. Richard Levenson, Professor and Vice Chair for Strategic Technologies in the Department of Pathology and Laboratory Medicine at UC Davis, spoke about how AI technology isn’t as infallible as it may seem. But first, he talked about another type of AI: Avian Intelligence, AKA pigeons.

Levenson and his colleagues published a study (DOI: 10.1371/journal.pone.0141357) in 2015 showing that in 15 days, pigeons could be trained to identify breast cancer on slides with an accuracy of about 80% per bird (the accuracy increased if more than one bird worked together). Similarly, he said, you can show a two-year-old child three or four pictures of cats, and they can distinguish it from dogs. Unlike a warm-blooded creature, AI still needs to see millions of images before they can make a distinction between cancer and not cancer, or cat and dog. In this way, though AI may ultimately be more precise, they take much more training than humans, or even pigeons. Levenson also said that in all the surrounding hype of AI, we shouldn’t forget that “old-fashioned” neural networks still work quite well in most clinical settings. “We don’t always need the biggest guns in the arsenal to do what we need to do,” he said.

And for all of AI’s solutions, Levenson said, they may also come with challenges that we haven’t yet fully considered. If AI is involved in an erroneous diagnosis, who is responsible for malpractice claims, Levenson asked. When considering replacing or enhancing physicians with AI, Levenson said, many more questions need to be addressed than simply which is more accurate.

On Wednesday, the AI debate resumed one last time. Keith Kaplan, CMO of Corista, called AI “the fourth industrial revolution,” but he seemed to be more moderate than the speakers on Tuesday. Kaplan neither praised AI nor disparaged it, instead taking a measured and practical approach to how AI can help clinicians. AI applications of the future could include more than just pathology, Kaplan said. Physicians could have AI assistants, there could be robotics and AI that helps seniors with independent living, and ultimately AI may save millions of dollars in the healthcare industry. But, Kaplan said firmly, AI in a clinical setting should be used with pathologists, “not in lieu of them.” Man and machine alone are not as effective as man and machine together, he said. The bottom line is that “the future is not AI versus MD, but AI plus MD.”

* The Molecular Medicine Tri-Conference; February 11-16, 2018; San Francisco. The Tri-Conference is produced by Cambridge Healthtech Institute, the parent company of Diagnostics World News.