Welcome to OpenForum. Join our global community: Sign up, Login or Learn more

Using Patient-Specific Cancer Mutanome and Pharmacogenomic Data to Create Personalized Avatars for Predicting Drug Efficacy and Safety

Patient-specific avatars for predicting predict drug response, discovering new indications for older drugs, and forecasting new drugs.

Photo of Christopher Cogle
5 1

Written by

Focus of Innovation

  • Collaboration
  • Cost reduction
  • Data collection
  • Data interpretation
  • Monitoring or measurement
  • Patient-centric or decentralization
  • Patient selection
  • Regulation
  • Research bias
  • Size of study
  • Target identification

Focus of Innovation - Other

Biomarker discovery, expansion of older drug labeling, new drug discovery, drug development

Name of applicant

Christopher R. Cogle, M.D.

Title of applicant

Associate Professor of Medicine, Oncology
Scholar in Clinical Research, Leukemia & Lymphoma Society

Affiliation of applicant

University of Florida

Your social media links


Names, titles and affiliations of teammates

Taher Abbasi, M.S., M.B.A.
CEO, Cellworks Group, Inc.

The greatest challenge in oncology is refractory cancer. One-third of all cancer patients die of refractory disease. In my blood cancer patient population, 80% ultimately die of cancer. Over the last 5 years, next generation DNA sequencing has unveiled three important characteristics of cancer: (1) high number of genomic mutations (also called mutanome), (2) highly heterogeneous mutanomes among cancers (even within the same tissue type), and (3) subclonal architecture with survival-of-the-fittest clonal evolution. These features explain the disparate disease responses in cancer patients and the lack of cure with one-size-fits-all treatment in many patients.

This challenge forged a partnership between an academic medical center and a computational biology company to invent new trials processes for introducing precision diagnostics and therapies to patients and the market.

The computational biology method includes software code based on tens of thousands of PubMed references that simulate the intracellular activities of thousands of known biological entities. Furthermore, code is written to reproduce tens of thousands of interactions among the known biological entities. Each patient’s cancer mutanome, which typically includes hundreds of gene mutations, is inputted into the computational biology system. A protein network map is then generated based on each patient’s cancer mutanome. Consequent mutant proteins are modeled as gain-of-function/activated or loss-of-function/inactivated. The patient’s cancer model/avatar is then used to quantify changes in cell growth characteristics (proliferation, viability and apoptosis) are quantified using growth curves. Drugs of interest are applied to the model and examined for reductions in cancer cell growth. Drugs with dose-dependent decreases in cell growth are predicted to remit the cancer.

The chief technology innovations in our approach are (A) the ability to include hundreds to thousands of gene abnormalities and (B) the ability to map the interacting nature of the multitude of gene abnormalities. This is the first technology capable of performing these tasks and in the clinic. These innovations generate new knowledge for understanding cancer biology, patient selection, and drug activity.

This method has been validated in several retrospective studies (Medina, et al. American Society of Hematology 2016 oral presentation). These studies have led to greater understanding of blood cancers and identification of unique biomarker signatures associating with drug response. Based on retrospective proof, we initiated a prospective validation trial called iCare for Cancer Patients (https://clinicaltrials.gov/show/NCT02435550). This study is powered to report predictive values of the computational biology method in predicting response to chemotherapy of choice. A follow-up prospective clinical trial (called iCare 2) is designed as a randomized phase 2 trial and is now under review at the FDA (Figure). These studies are necessary to validate the clinical accuracy and utility of a computational biology method. To our knowledge this is the only precision oncology computational biology method in randomized testing for validation.

From this effort, several innovative reinventions of the trials process have arisen:

  • A rapid workflow for N=1 treatment decision-making in cancer,
  • Rapid method of repurposing older drugs for new indications. These discoveries joined with clinical proof-points help pharmaceutical companies reinvigorate older assets,
  • Ability to identify novel biomarker signatures that explain drug sensitivity or resistance. These biomarker signatures not only further our understanding of cancer biology and pharmacology, but also represent patentable intellectual property (IP) for creation of companion diagnostics,
  • A virtual clinical trials method using large, publicly available datasets of cancer patients to test clinical efficacy of investigational new drugs. Drugs generating improvement in virtual Kaplan-Meier survival curves are fast-tracked for human application. Drugs generating no improvement in Kaplan-Meier survival curves are scrutinized further with other pre-clinical examinations. In general, the earlier an investigational agent is identified as a poor performer, the better it is to conduct further structure activity relationship (SAR) studies or scuttle the effort and focus on more promising assets.
  • Rapid pharmacogenomics screen for predicting drug metabolism and adverse events. These data help prepare the patient and treating physician for what to expect. Prophylactic measures such as enhancing anti-emetic strategy can be employed. These pharmacogenomics screening profiles represent new IP that for companion diagnostics with every drug.
  • Monitoring cancer burden by saliva or finger stick. Non-invasive specimens from cancer patients can be measured rapidly (within days) for cancer burden using ultra-high sensitivity droplet digital polymerase chain reaction (ddPCR). This method not only permits rapid determination of biologic activity, but also permits early remodeling and readjustment of anti-cancer treatment strategy. There is a growing agreement among oncologists that, in the future, refractory cancers will be managed by a myriad of drugs based on cancer subclonal evolutions.

Why would your idea have a significant impact?

The mission of our partnership is to cure patients with cancer.

What are the biggest hurdles to implementing your idea?

Regulatory approvals from the FDA, SFDA and EMA are required for personalized medicine methods with significant risk to patients. Treatment decision-making is considered a significant risk. Few guidelines exist for computational devices used for personalizing therapies.

Repurposing drugs is within the legal and medical purview of oncologists for treating patients with cancer. In fact, most drugs used in oncology are off-label. However, repurposing newer drugs is costly and health insurance companies may not cover those costs.

How difficult will it be to overcome those obstacles?

In regards to regulatory approvals, to mitigate risk to the patient, we have instituted several safety measures and checks-and-balances, including among others a molecular oncology board to serve as a filter between the computational output and the patient.

In regards to drug coverage of off-label drugs for cancer, 1993 federal legislation requires insurance companies to cover medically appropriate cancer therapies. In 2008, Medicare rules were changed to cover more off-label uses of cancer treatment drugs. One criterion for drug coverage is peer-reviewed medical literature supporting the drug’s use in cancer. We have included the PubMed reports used in creating each patient’s cancer model and simulating drug efficacy. These reports are provided to the health insurance company as evidence of medical necessity. Ultimately, we believe that health insurance companies will find great value in this technology, as it will enable early identification of responders versus non-responders and prompt digital drug screening for most efficacious drug treatment.

How would you test for the impact of this idea and how would you quantify the impact?

After demonstrating 80-100% accuracy in matching drug response, we are now conducting a prospective validation study (https://clinicaltrials.gov/show/NCT02435550). A positive predictive value of 70% or greater would indicate high clinical utility. Several phase 2 clinical trials of patients with model-discovered unique biomarker signatures are in design now. We expect at least a doubling of response rates (e.g., 60% vs. 30%) compared to cancer patients without the unique biomarker signatures. The use of unique biomarker signatures generating higher rates of response enables smaller sample sizes, which quickens the pace towards drug approval.


Join the conversation:

Photo of Anne Cherry


Thank you for sharing this very interesting idea. Reading it, it seems that you and Andrew Beam  are working on highly similar projects – he is documenting the gene expression signatures of cancer cells while you are documenting the genomic signatures of cancer cells – both with an eye towards identifying the most appropriate therapies for a given patient.

I have two broad questions about your submission: Firstly, how will you generate a computational platform that can accurately predict all the variables in such a vastly complex system? Secondly, if this computational platform can be created to predict all these things, how do you expect its use will be extended into clinical trials?

If I understand your proposal (and please correct me if I’m mistaken), the computational platform must be able to make a number of accurate predictions in order to provide advice regarding the most appropriate therapy for a given tumor:

1) Identify mutations in cancer cell genomes (presumably by comparing to non-cancer patient tissue)

2) Predict whether each coding mutation causes gain-of-function, loss-of-function, or unchanged function (e.g., passenger mutation)

3) Predict whether each non-coding mutation influences protein levels, and in what way(s)

4) Predict the protein network resulting from that mutanome (including expression rates, transcription efficiency, protein localization, and degradation rates)

5) Predict how this protein network will influence cell growth and death patterns

6) Predict how a given drug will influence the protein network

7) Predict how this change will impact cell growth and death patterns

Any single one of those steps may require hundreds (thousands? millions?) of experiments in order to generate sufficient data to make these predictions accurately – but an experimental shotgun approach does not seem feasible, at least in the short term. What is your group’s methodology to help the computational platform make accurate predictions, even in the absence of data?

My second question relates to how this platform could be employed in a clinical trial system. It sounds like the primary goal is to determine the most effective therapy for a given patient – what do you anticipate its role would be in a clinical trial setting? Particularly, given that different cell types and different types of tumor (not to mention different mutanomes) respond differently to stimuli, how will you predict the impact of a never-before-used drug on a given tumor’s protein network? Would you need to test each potential trial drug on lots of different samples (e.g., cultured cancer lines of different types, cultured non-cancer lines, primary cell types of tumor cells-of-origin, other patient samples) in order to accurately determine whether a given patient’s tumor would be likely to respond to that drug? I suppose I’m struggling with what seems like a bit of a disconnect between setting up the system (steps 1 – 5 above), and then how it will actually perform virtual trials. How will you know which proteins a given drug will impact, if it’s never been put into a cell before?

I think it’s very exciting that your group is already starting to implement this kind of computational system. I wish you all the best!



View all comments