NEW YORK – Utah Science Technology and Research Initiative (USTAR) spinout Eigengene has developed a proprietary mathematical approach to discover patterns in cancer whole-genome sequencing data. Headquartered in Palo Alto, California, the firm believes it can use its technique to build clinically actionable genetic signatures in a wide variety of cancers for personalized prognostic, companion diagnostic, and therapeutic purposes.
Last month, the researchers behind Eigengene published a proof-of-principle study in APL Engineering that highlighted the method's ability to predict the survival of glioblastoma multiforme (GBM) patients based on a genome-wide pattern of copy number alterations in their tumors. The prediction was better than standard methods, which involve the patient's age at diagnosis.
Orly Alter, an associate professor of bioengineering and human genetics at the University of Utah and the senior author of the study; Sri Priya Ponnapalli, a senior research affiliate in Alter's lab and the study's lead author; and David Oberman, Alter's husband, cofounded Eigengene in 2016 to commercialize mathematical models developed in Alter's lab — including the one in the paper, which uses comparative spectral decomposition — for the clinical space.
Alter explained that comparative spectral decomposition collects multiple datasets and separates them into patterns, which "essentially describe everything that happens in the datasets." The method then examines what is common between and exclusive to either one of the datasets.
By applying comparative spectral decomposition to DNA copy number variants in a patient's tumor and normal tissue, Alter said that she and her colleagues identified genome-wide patterns for multiple cancer types, such as glioblastoma and lung adenocarcinoma.
To do that, the researchers first computed whole-genome sequencing (WGS) profiles for tumor tissues and blood samples of patients, which Alter said requires about five minutes to for dozens of patients. They also repeatedly ran several iterations — anywhere from hundreds to thousands of times — to improve the robustness of the pattern by making small alterations each time, Alter said, which can take up to several months.
However, once a pattern has been discovered, validated, and tested, Alter said the process to classify a cancer patient based on the pattern and their genome sequence "only requires minutes."
"[This enables] the separation of patterns [that] occur only in the tumor genomes from those that occur in the genomes of normal cells in the body and are conserved in the tumor cells," she added.
Alter's team initially discovered GBM genome-wide patterns of DNA copy number alterations by comparing normal and tumor Agilent microarray profiles of 251 GBM patients from the Cancer Genome Atlas (TCGA). The group then validated the pattern using tumor TCGA Agilent profiles from an independent set of 184 TCGA GBM patients.
In the proof-of-principle study, Alter's team performed WGS in two duplicate sets of DNA samples extracted from tumor samples of 79 GBM patients. The group then analyzed the WGS tumor profiles using comparative spectral decomposition and found that the GBM pattern predicted patient survival better than age at diagnosis, and independently of it.
While acknowledging that the study is both interesting and hypothesis-generating, Ingo Mellinghoff, co-chair of Memorial Sloan Kettering Cancer Center's neurology department, who was not affiliated with the study, noted that future prospective studies will be needed to determine the prognostic value of the GBM copy number pattern.
"What is most urgently needed in neuro-oncology are biomarkers that can predict response to specific therapies (i.e. 'predictive') biomarkers," Mellinghoff said in an email. "Prognostic biomarkers can only take you so far, in particular given the current dearth of effective alternative treatments for GBM."
While in the study, Alter and her colleagues validated a signature that predicts life expectancy of GBM patients, she also noted that her team has discovered and validated genome-wide patterns that predict the survival of patients for other brain cancers, including lower-grade astrocytoma and neuroblastoma. The patterns for lower-grade astrocytoma are the same as the GBM patterns from the study, Alter said, but neuroblastoma genome-wide patterns are different, and her team is currently preparing a separate study on that.
Alter said her team has also validated signatures that predict the survival of lung, ovarian, and uterine adenocarcinoma patients in response to platinum-based chemotherapy, and the researchers are planning to analyze data from breast cancer and melanoma.
Eigengene now aims to bring the models Alter's group developed for GBM into the clinical space. The firm will also scale up its commercial efforts by looking into the signatures for lung, ovarian, and uterine cancers.
"We're looking at cross-cancer comparisons and mechanisms that may be universal in cancers," Alter said. "We're considering data from liquid biopsy, imaging, and even digital health, [and thus are] also interested in joining efforts for developing novel therapeutics and companion drug diagnostics."
To that end, Eigengene aims to partner with pharmaceutical companies on end-to-end AI-driven drug development, as well as with cancer centers and genomic sequencing companies to revalidate the signature patterns in retrospective and prospective trials.
"Our goals are to increase drug approval rates, help existing drugs get US Food and Drug Administration approval for off-label uses, accelerate new molecules from Phase I clinical trials to FDA approval, and discover novel targets and drugs," Alter said.
However, she declined to disclose information about the firm's growth since it launched in 2016, any intellectual property the firm has filed for in relation to comparative spectral decomposition, and current licensing agreements with the University of Utah.
The team also declined to say how the company is funded. It is not seeking additional funding at the moment, Alter noted.