Skip to main content
Premium Trial:

Request an Annual Quote

Applied Informatic Solutions Uses Modeling to ID Predictive Biomarkers for Cancer Treatment


Applied Informatic Solutions is hoping that its mathematical models and informatics services will attract industry partners looking to develop prognostic tests and targeted drugs.

The three-person, St. Paul, Minn.-based company opened its doors a year and a half ago and has developed a series of mathematical models that it claims can identify prognostic biomarkers in breast and ovarian cancer that can predict response to standard chemotherapy treatments.

Paul Burgio, the company's president and co-founder, told BioInform this week that the company has submitted patent applications for both its ovarian and breast cancer prognostic models.

If the patents are granted, the company would be able to sell licenses for the right to use its patented technology to develop prognostic tests for ovarian and breast cancer, Burgio said.

Alternatively, a drug developer looking to create a drug for either kind of cancer would likely be interested in using the models to select for clinical trials patients who would best respond to a proposed therapy, he said.

AIS' models are based on a mathematical concept called the theory of super variables, which was developed by company co-founder Jason Nikas. The concept is used to reduce the dimensionality of complex data sets by first identifying an initial pool of significant variables and then using mathematical modeling to generate a single super variable that can discriminate between the control and the experimental groups with very high accuracy.

Nikas explained to BioInform that the company fits mathematical functions based on this theory to biological datasets in order to find the best possible explanations for phenomena that are being observed.

AIS recently published a paper in Biomarker Insights in which it discusses its prognostic biomarker model for breast cancer that predicts which patients will respond to paclitaxel-fluorouracil-doxorubicin-cyclophosphamide, or T/FAC, treatment.

In that paper, the authors explain that they developed the breast cancer model using gene expression information from excised tumors of 50 patients — 10 who responded to treatment and 40 who didn’t — and then they validated the model on a separate dataset obtained from analysis of 43 new patients — 10 of whom responded and 33 who didn’t respond to treatment.

According to the paper, the researchers first identified 14 significant genes, of which nine were used as the final super variable. These nine genes are involved in transcription regulation; cell proliferation, invasion, and migration; oncogenesis; suppressed immune response; drug resistance; and cancer recurrence.

Based on over- or under-expression of these genes — which include CCDN1, LAMA5, and RARA — the model is able to predict which breast cancer patients will respond to chemotherapy with a sensitivity and specificity of 90 percent and 92 percent respectively, the authors wrote.

A second paper, published in Cancer Informatics last October, describes three separate sets of super variables for ovarian cancer that predict which patients will respond to the standard platinum/taxol chemotherapy treatment used after surgery.

The three models were derived from an initial set of 84 significant genes and comprise a total of 13 genes, of which five are in all three models. The three function "completely independently" of each other, and each one "could serve as a prognostic test by itself [and] it could be licensed to one company," Nikas told BioInform.

It would also be possible to put all three functions into a software package along with a decision tree algorithm that would predict a particular treatment result based on concurring results between two of the models, he said.

The models were developed based on data from 34 patients and validated with 20 new subjects. They predict response to treatment with sensitivity and specificity that ranges from about 96 percent to 100 percent.

In both cancer cases, the input genes were selected using several statistical significance tests, such as p-values and ROC curve analysis, to derive a pool of likely prognostic gene candidates, Nikas explained.

According to both papers, there is a dearth of methods for testing treatment response in breast and ovarian cancer.

While there are several gene expression tests available to predict the likelihood of breast cancer recurrence, such as Agendia's MammaPrint and Genomic Health's Oncotype DX, these tests are currently not approved to predict response to specific treatments. The Biomarker Insights paper states that there are "currently no biomarkers available that can predict which breast cancer patients will respond to chemotherapy with both sensitivity and specificity [greater than] 80 percent, as mandated by the latest [US Food and Drug Administration] requirements."

Similarly, the Cancer Informatics paper states that currently there are no prognostic tests that, "at the time of the diagnosis/surgery can identify those patients with advanced stage epithelial ovarian cancer who will respond to chemotherapy."

The need for such a test is particularly evident in ovarian cancer where more than 60 percent of cases are diagnosed when the disease is already at an advanced stage, "due to lack of early symptoms," the Cancer Informatics paper states.

At that point, the five-year survival rate of those patients drops to less than 30 percent, the paper said.

Furthermore, once the disease has been diagnosed and chemotherapy begun, 70 percent to 80 percent of patients initially respond to the platinum/taxol chemotherapy, but more than 75 percent of them soon experience a recurrence, the paper states.

Armed with the knowledge of their possible response or non-response to particular drugs, patients and their physicians can look into alternative treatments earlier in the process and, by extension, improve their chances of survival, Nikas said.

In addition to its work on breast and ovarian cancer models, the company is applying its mathematical functions to find biomarkers for other diseases such as colorectal cancer, Burgio told BioInform.

He said that the company plans to work with customers to develop models that are tailored to work specifically on their datasets and that it is currently talking to potential clients but he declined to provide further details since negotiations are ongoing.

AIS charges an undisclosed initial analysis fee based on the size and complexity of the dataset in question, Burgio explained. If the company is able to identify useful biomarkers with high sensitivity and specificity, then it charges a second fee to transfer the technology and implement it at the customer's site.

Furthermore, although AIS expects the bulk of its business to come from pharmaceutical and diagnostic development companies, the company is open to applying its mathematical models in the environmental and physical sciences as well as in energy, he said.

Filed under

The Scan

Mosquitos Genetically Modified to Prevent Malaria Spread

A gene drive approach could be used to render mosquitos unable to spread malaria, researchers report in Science Advances.

Gut Microbiomes Allow Bears to Grow to Similar Sizes Despite Differing Diets

Researchers in Scientific Reports find that the makeup of brown bears' gut microbiomes allows them to reach similar sizes even when feasting on different foods.

Finding Safe Harbor in the Human Genome

In Genome Biology, researchers present a new approach to identify genomic safe harbors where transgenes can be expressed without affecting host cell function.

New Data Point to Nuanced Relationship Between Major Depression, Bipolar Disorder

Lund University researchers in JAMA Psychiatry uncover overlapping genetic liabilities for major depression and bipolar disorder.