Skip to main content
Premium Trial:

Request an Annual Quote

Helio Genomics Mulls New AI-Driven Dx Opportunities While Readying Liver Cancer Test for US

Premium
DNA methylation

NEW YORK – Helio Genomics, a company that has already used artificial intelligence tools to develop a liver cancer test, has ambitions to develop other precise cancer diagnostics by applying deep learning models and generative AI.

In Genome Medicine in January, Helio Genomics, formerly known as Helio Health, published results from a study assessing the effectiveness of an AI-driven platform dubbed MESA, which stands for multimodal epigenetic sequencing analysis, as a colorectal cancer screening tool. The platform combines cell-free DNA (cfDNA) methylation analysis with machine learning methods to identify biomarker signatures that can detect cancer in its earliest stages. MESA is designed to detect cancer early by incorporating four parameters: cfDNA methylation, nucleosome occupancy, nucleosome fuzziness — a measure of delocalization of the nucleosome — and a windowed protection score for regions surrounding gene promoters and polyadenylation sites.

In the study, researchers from Helio and the University of California, Irvine's computational biomedicine division analyzed 690 cfDNA samples from three colorectal cancer cohorts, including samples collected from patients in the ELITE trial, held by contract research organizations, and samples housed at Sun Yat-sen University Cancer Center in Guangzhou, China. Helio is conducting the observational ELITE trial to assess the performance of a MESA-based test for detecting cancer and to discover new biomarkers for various cancer types in about 1,200 patients with and without cancer.

After processing the samples, sequencing, and extracting information on the four features, investigators developed a multimodal machine learning model for cancer detection using samples from two out of the three patient cohorts and performed cross-cohort validation with the third. The researchers found that cfDNA methylation and nucleosome occupancy/fuzziness enabled accurate detection of colorectal cancer with MESA, but integrating all four features yielded the highest sensitivity with an area under the curve of 0.88.

This colorectal cancer test is Helio's latest application of AI to develop a diagnostic test, though use of the technology is integral to all its projects, including its flagship product, HelioLiver.

Helio CEO Justin Li said AI technology was critical to handling the large amount of data involved in developing this test. "There's a pattern that exists in biological data that holds the secret to whether a patient has cancer or not," Li said. "AI is the only reason that anybody in our industry can make sense of that information."

HelioLiver is a MESA-based test that interrogates 77 methylation sites and three proteins. The test is currently offered through a commercialization and co-branding partnership with Fulgent Genetics. Helio developed the test using the ECLIPSE platform, an automated process that combines the steps involved in extracting and capturing cfDNA with next-generation sequencing into a single workflow.

Liver cancer surveillance is typically carried out via serum alpha-fetoprotein (AFP) testing and ultrasound. The sensitivity of ultrasound alone for detecting early-stage liver cancer is only 45 percent. When AFP testing is added, the sensitivity increases to just 63 percent. Ultrasound results also vary widely by operator and by patient body mass index.

In a validation study in 247 patients with liver cancer, Helio researchers compared the performance of HelioLiver to both AFP testing and the established GALAD score, which is calculated from age, sex, and several biomarkers including AFP. HelioLiver was superior to both in detecting early-stage liver cancer, and they reported a sensitivity of 75.7 percent and a specificity of 91.2 percent for the test.

The West Lafayette, Indiana-based company has three clinical trials of HelioLiver that are ongoing or completed. In the prospective Phase II ENCORE study, the company evaluated the performance of the test in patients with stage I through stage IV hepatocellular carcinoma, patients with other cancers, and control subjects without cancer in China. In 2022, Helio published results from ENCORE in Hepatology Communications that showed that HelioLiver had 91 percent specificity and 76 percent sensitivity for detecting early-stage liver cancer.

In the US-based CLiMB study, the company is comparing the sensitivity and specificity of HelioLiver with ultrasound in about 1,600 patients with high-risk liver cancer due to liver cirrhosis. And in the LIVER-1 study, Helio aims to validate the test in a population at high risk for hepatocellular carcinoma due to liver disease. That study is also being conducted in the US.

A test for liver cancer is of particular interest in Asia, especially in China and Korea, where the prevalence of the disease is high. For example, one study estimated that in 2020, there were 46,600 primary liver cancer cases and 34,800 deaths in North America, compared to 99,300 cases and 95,700 deaths in Southeast Asia, or 491,700 cases and 449,500 deaths in Eastern Asia.

"With the data that we saw in China, we were quite confident that [Helio would also be able to] generate positive data in the US," said Sunwoo Kim, a partner and portfolio manager for healthcare at Seoul, South Korea-based Quad Investment Management. 

Kim said his firm invested in Helio primarily because of the "superior results" seen to date in trials of HelioLiver and the management team's track record at previous startups.

Helio is now preparing to submit data from the CLiMB trial to the US Food and Drug Administration as part of its bid to garner approval of HelioLiver as an in vitro diagnostic. Helio is also planning to "invest heavily" in commercializing the test, getting it established in treatment guidelines, and securing payor reimbursement for it, Li said.

In anticipation of a potential US launch for the test, Helio is working with hospitals and doctors to define its intended-use population and where it could fit in the medical workflow. "We've already finished the clinical trials, so we're essentially ready to market and get that product out there into physicians' hands," Li said.

The company is also investigating applications of its MESA technology in indications beyond liver cancer such as colon, stomach, and lung cancer. The indication Helio pursues next, according to Li, will be influenced by whether a clinical model for use of such a test already exists. For example, in liver cancer, patients are already receiving ultrasound screening every six months, and positive results are confirmable via MRI or CT imaging and are actionable in terms of treatment. "We want to find that same [model] for our next cancer type," Li said, adding that though plans are preliminary, there are indications that similar frameworks may exist in lung cancer and stomach cancer.

Li said AI will continue to play a significant role in Helio's future plans. Alongside its product development strategy, Helio is planning to double down on its use of AI by expanding the scope of its technology to include deep learning and generative AI.

"Deep learning is the next stage of the company," Li said, noting that a machine learning algorithm is trained on labeled data, whereas a deep learning model does not. For example, he explained, in training a machine learning model to detect cancer, the patient data would be labeled, indicating which patients have cancer and which do not. In contrast, a deep learning model is more comparable to a large language model such as ChatGPT, and the algorithm would sift through unlabeled patient data for patterns.

Li said Helio's scientists have been working with deep learning models for more than a year, and the company has invested in servers and graphics processing units as computational resources. "We've seen some encouraging data internally," Li said. "But we want to test out different models and different populations before we make any announcements as to the efficacy of deep learning in our space."