Harnessing Supercomputers to Improve Cancer Immunotherapy

Latest efforts help scientists identify the efficacy of immune treatments in patients with cancer.

Immunotherapies have created waves in the oncology landscape, but responses to treatment vary from patient to patient. While some may have a positive response to treatment, others may experience powerful, sometimes even fatal, adverse events.

More knowledge needs to be learned on exactly how the immune system fights cancer, and supercomputers may play a key role in uncovering this information.

Scientists developed a novel mathematical model to examine the interactions between prostate cancer tumors and common immunotherapy approaches.

In a study published in Scientific Reports, they used the model to predict how prostate cancer would react to 4 common immunotherapies: androgen deprivation therapy, vaccines, Treg depletion, and IL-2 neutralization.

The investigators harnessed computer simulations and data analysis to study the systemic effects of the 4 immunotherapies. They incorporated data from animal studies into their mathematical models and simulated tumor responses to treatments using the Stampede supercomputer at the Texas Advanced Computing Center (TACC).

“We do a lot of modeling which relies on millions of simulations,” said investigator Jing Su. “To get a reliable result, we have to repeat each computation at least 100 times. We want to explore the combinations and effects and different conditions and their results.”

The results of the study showed that Treg depletion and IL-2 neutralization had higher efficacy when combined with androgen deprivation therapy and vaccines.

The findings suggest the potential therapeutic strategy that could help improve the management of prostate tumor growth, according to the authors.

In a separate study published in April of this year, investigator Xiaobo Zhou used high performance computing resources to predict how RNA and proteins interact with greater accuracy than previous methods.

The investigators conducted an analysis of 1342 RNA-protein interacting complexes from the Nucleic Acid Database to identify diverse interface properties between them, which included binding and non-binding sites.

Next, the investigators used a 3-step method to predict the interacting regions between the RNA and protein using both their sequences and structures. The results of the study showed that the model was more accurate and outperformed the leading existing method by 20%.

“TACC provides an important assistance for discovering clinically meaningful and actionable knowledge across high heterogeneous biomedical big data sets,” Zhou said.

Novel Dose-Finding Designs That Consider Toxicity and Efficacy of Biological Agents in Need

The function of biological agents in immunotherapy differ from chemotherapy and radiation. Typically, toxicity and efficacy increase with the dose level of cell-destroying chemicals or X-rays, but this may be different for biological agents.

Toxicity may increase at low dose levels, and then plateau at higher dose levels when the biological agent reaches a saturation level in the body. Additionally, efficacy may even decrease at higher dose levels.

Traditional dose-finding designs focus on identifying the maximum tolerated dose that is unsuitable for biological agent trials. Therefore, designing trials that consider both the toxicity and efficacy of these agents is crucial.

Scientist Chunyan Cai uses TACC systems to design novel dose-finding trails for combinations of immunotherapies. In a paper published in the Journal of the Royal Statistics Society Series (Applied Statistics), Cai and her colleagues described their efforts to identify biologically optimal dose combinations (BODC) for agents that target the PI3K/AKT/mTOR signaling pathway.

“Our research is motivated by a drug combination trial at the MD Anderson Cancer Center for patients diagnosed with relapsed lymphoma,” Cai said. “The trial combined 2 novel biological agents that target 2 different components in the PI3K/AKT/mTOR signaling pathway.”

Individually, both agents partially inhibit the signaling pathway and provide therapeutic value. The investigators hypothesized that combining the 2 agents would result in a more complete inhibition of the PI3K/AKT/mTOR signaling pathway.

In the study, the investigators examined the combinations of 4 dose levels of agent A with 4 dose levels of agent B, resulting in 16 dose combinations, to identify the biologically optimal dose combination.

The investigators introduced a dose-finding trial design that accounted for the unique properties of biological agents, according to the study.

“Our design is conducted in 2 stages,” Cai said. “In stage 1, we escalate doses along the diagonal of the dose combination matrix as a fast exploration of the dosing space. In stage 2, on the basis of the observed toxicity and efficacy data from stages 1, we continuously update the posterior estimates of toxicity and efficacy and assign patients to the most appropriate dose combination.”

Six different dose-toxicity and dose-efficacy scenarios were examined, and the investigators carried out 2000 simulated trials of each design using the Lonestar supercomputer.

The simulations compared the percentage of the BODC, percentage of patients allocated to the BODC, the average efficacy rate, number of patients assigned to over-toxic doses, and the total number of patients assigned in stage 1 and 2 of the trial.

The investigators found that the optimal dose-finding design gives higher priority to trying new doses in the early stage of the trial, and assigned patients to the most efficacious and safe dose towards the end of the trial.

“Extensive simulation studies show that the design has desirable operating characteristics in identifying the biological optimal dose combination under various patters of dose-toxicity and does-efficacy relationships,” Cai concluded.

Encouraging Data Sharing Among Research Community

Mechanisms to share, compare, and integrate disparate research findings among the scientific community is critical.

To address this need, the VDJServer was launched last year to enable researchers to analyze high-throughput immune repertoire sequencing data over the web using the high-performance computing resources available at TACC.

Repertoire sequencing has helped transform the immunotherapy landscape by enabling quantitative analyses that help scientists understand the function of immunity in health and disease.

“VDJServer provides access to sophisticated analysis software and TACC’s high-performance computing resources through an intuitive interface designed for users who are primarily biologists and clinicians,” said project leader Lindsay Cowell. “In addition, we provide platforms for sharing data, analysis results, and analysis pipelines. Access to these analyses and resource-sharing accelerates research and enables insights that wouldn’t be possible without the opportunity for data integration.”

To perform data-driven studies, investigators can upload B and T cell receptor data and harness the TACC’s computing power through the site.

“Immunotherapy is a relatively young field and the computational tools are emerging alongside with knowledge of the domain,” said Matt Vaughn, director of Life Science Computing at TACC. “Community-oriented efforts like VDJServer are important because they provide a centralized workbench where best of breed algorithms and workflows can be used much more quickly than if they are released just as a source code and at the end of a long publication cycle. They’re also available democratically: anyone can use software at VDJServer regardless of how computationally experienced they are.”