Skip to main content
Premium Trial:

Request an Annual Quote

Philips, Case Western InFlo Software Uses Multi-Omics Data to Find Anomalous Signaling in Cancer

Premium

NEW YORK (GenomeWeb) – Scientists at the Case Comprehensive Cancer Center at Case Western University and their collaborators at Philips Genomics and Princeton University have developed new software, called InFlo, that combines omics data with pathway information to identify anomalies in cellular signaling networks that play a role in tumor development.

According to a paper published last month in Oncogene, the software estimates the activities of interactions within cell signaling pathways in individual patient samples by integrating information from multiple molecular measurements including DNA, RNA, proteins, and methylation data. The approach is tissue specific and includes strategies for reducing noise in the data as well as for highlighting errors or disruptions in pathway structures.

Researches can use the software to look at signaling differences between cancerous and healthy tissues and potentially identify markers that could serve as therapy targets, according to the developers. "Cellular signaling networks are the mechanisms that cells use to transfer, process, and respond to biological information derived from their immediate surroundings," Vinay Varadan, an assistant professor at Case Western Reserve University School of Medicine and one of the corresponding authors of the study, explained in a statement. "InFlo can be viewed as modeling the flow of information within these signaling networks."

Researchers at Philips Genomics plan to incorporate InFlo into the Philips IntelliSpace Genomics platform, a software solution that they have developed for combining and analyzing genomics and other kinds of data in both clinical and research contexts. In fact, Varadan began developing the software while he was employed at Philips Research, where he worked with Nevenka Dimitrova, the chief technology officer for genomics in the Philips Healthcare division. 

The central question the researchers sought to address was "if I only had one patient and I could measure multiple things, how would I find out what pathways or signaling networks are hyperactivated in that patient's tumor?" according to Varadan. To that end, "we were very interested in [methods of] integrat[ing] molecular profiling data on individual patient samples," he said, but at the time, there were no methods specifically designed for that purpose.

Varadan left Philips in 2013 for a position at Case Western, where he and other colleagues were able to complete much of the work on InFlo. They were also able to validate the software in the context of different cancers, including gastrointestinal, breast, and ovarian cancer. The Oncogene paper describes the results of applying InFlo to high-grade serous ovarian carcinoma samples from the Cancer Genome Atlas. The researchers used the method to identify signaling pathways that are associated with resistance to platinum-based chemotherapies.

Specifically, they identified an interaction between two proteins, cAMP and CREB1, as a key mechanism associated with platinum resistance. Furthermore, by following up on the predictions provided by their models, they found that inhibiting CREB1 sensitized ovarian cancer cells to platinum therapy and is also effective in killing ovarian cancer stem cells.

According to Analisa DiFeo, co-corresponding author of the InFlo study and professor of ovarian cancer research at Case Western's medical school, the researchers are currently evaluating whether this could be a therapeutic target for treating cases of platinum-resistant ovarian cancer.

Essentially, the software takes two different kinds of information and merges them together using a mathematical model, Varadan explained. The first bit of information is molecular profiling data, such as gene expression, copy number, and methylation data gleaned from a single patient's tumor. The software also incorporates curated signaling pathway networks from repositories such as the National Cancer Institute's Pathway Interaction Database, Reactome, and the Kyoto Encyclopedia of Genes and Genomes, which provide details on the interactions between different proteins, the directionality of these interactions, and the activation or deactivation status.

In a first step, the software uses probabilistic models to compare information from the diseased tissue to information from a small number of normal samples of the same tissue type from different patients and assesses the differences between the two. For example, the software would look at whether certain genes are upregulated or downregulated in the disease sample compared to the normal samples or whether there are more or fewer copy numbers, Varadan explained. These models calculate a probability that a gene is activated, deactivated, or neutral in a cancerous tissue.

The next step is to integrate the gene level information with signaling pathway information. "What InFlo does is, it first parses the networks that are there in these signaling network databases and extracts out these individual [protein] interactions within each signaling network," Varadan explained. It then models the protein complexes that are formed by interactions between proteins that likely play a role in tumor development. The software then uses a generative model to estimate the probabilities of the interaction activities in the tumor sample by combining the gene-level probabilities with the signaling pathway network-based interactions.

Furthermore, InFlo's models account for differences in the tissues used to study the signaling networks reported in the databases, Varadan added. Specifically, it performs a "consistency check" of the interaction activities that it has computed.

For example, if a given pathway is structured in a way that indicates that a downstream protein should be activated, the software checks whether that is true or not in the tissue sample being studied. This consistency check allows InFlo to increase the robustness of the interaction activity estimates on a per-tumor basis. Researchers can then visualize the results of their analysis in the Cytoscape software.

Varadan's team at Case Western is now working with Philips researchers to incorporate InFlo into the Philips IntelliSpace Genomics platform — Philips actually holds the patent on the core InFlo algorithm. Currently, "we are transferring the code to the development folks on the Philips side and designing wrappers around our code so that it can talk to their modules on their platform," Varadan said.

Philips is incorporating InFlo into the research-focused iteration of the IntelliSpace Genomics software, Dimitrova told GenomeWeb. IntelliSpace Genomics is part of a broader suite of Philips-designed healthcare products for capturing and integrating patients' healthcare and lifestyle data, including laboratory results, pathology images, and clinical history.

The research version of the software offers computational workflows for analyzing genomic and phenotypic data, including supervised and unsupervised machine-learning methods for identifying genes that behave differently in different conditions. There are also tools for analyzing disease pathways and for visualizing genomic data.

"Within that context now [researchers] can use Inflow to generate hypotheses," she said. For example, they can search for signaling pathways that are responsible for the dysregulation observed in data from two sample groups, such as tumor versus normal or response to therapy versus non-response. It will take a long time, however, before the fruits of InFlo could be used in the clinical space, she said. For example, the interaction reported in the study between cAMP and CREB1 and the role it plays in chemotherapy resistance would need to be validated  in multiple studies. "We show that this is true in a cell line but ... in order for somebody to use this in clinical practice, there are many steps on that road."

Meanwhile, the core InFlo algorithm will remain free for academic research use. Varadan and his colleagues are already working on new features for the software, including expanding it to incorporate other kinds of data. For example, they hope to be able to integrate additional types of epigenetic data into the molecular networks. "DNA methylation we've already shown would be possible but we are also talking about other kinds of epigenetic variations in these tissues that we can measure, [such as] gene fusions," he told GenomeWeb.

They are also collaborating with the imaging informatics research group within Case Western's Center for Computational Imaging and Personalized Diagnostics to integrate imaging features derived from pathology and radiology data into InFlo. The initial focus of that collaboration will be breast cancer but "we have a wide number of cancers that we are going after," Varadan said.

Other plans include enabling InFlo to identify unknown players in molecular networks. "We find there are inconsistencies that keep showing up over and over again in a particular tissue context," Varadan said. The planned updates will make it possible to ask the question "is that happening because of mutations in some of the genes, or is that happening because there's a novel microRNA or lincRNA involved?" he said. "We are developing InFlo to be able to handle that."

Philips recently announced two partnerships around IntelliSpace Genomics that will focus on personalizing cancer treatments. One of these is with the Westchester Medical Center Health Network, which is using IntelliSpace to combine next-generation sequencing and clinical data to offer more tailored treatments for patients. The second is a collaboration with clinical interpretation services provider N-of-One, aimed at enhancing the capabilities of the IntelliSpace Genomics solution.