Tumor Neoantigen Selection Alliance (TESLA)

Why This Research

Tumor neoantigens act like a red flag to the immune system: they tell the body something has gone wrong. In this case, that something is cancer. Growing out of control, mutating.

Scientists figured out a way to take neoantigens and turn them into targets for personalized immunotherapies to each patient.

But how can we predict which mutations in a patient’s DNA will make the best personalized therapeutic? One that helps kill cancer cells but not normal cells, and riles up the immune system in just the right way?

You take the best minds, the best algorithms and form an AI cancer supergroup to find the answer.

What We’re Doing

PICI and the Cancer Research Institute (CRI) did just that when they launched the Tumor Neoantigen Selection Alliance in fall 2016. This global bioinformatics consortium includes scientists from 36+ of the leading neoantigen research groups in academia, nonprofit and industry.

Through predictive algorithms and machine learning, the group is sniffing out which cancer neoantigens encoded in DNA can be recognized and stimulate an immune response.

Finding the right predictive algorithms for targeting neoantigens could allow scientists to create more cancer immunotherapy treatments tailored to each patient.

About the Research

Research teams took DNA sequences from normal, melanoma and non-small cell lung cancer (NSCLC) tissues. Using their own algorithms and machine learning methods, they formulated a list of predicted neoantigens anticipated to be present on the tumor cells and recognizable by the immune system.

To see how well those predictions stacked up, the proposed neoantigens were evaluated through laboratory testing.

Each participant received feedback to inform and improve their predictions in the future. With each iteration, predictions should become more precise.

This data-driven process will help us pinpoint the best targets for personalized cancer treatments. The end result: more effective, targeted cancer vaccines or cell therapies tailored to each patient.

Results

Findings published in Cell on October 9, 2020, showed that no team’s methodology identified every neoantigen, nor a large majority of these cancer markers, indicating a need for a harmonized scientific effort like TESLA.

More significantly, the alliance discovered five characteristics that strongly indicated which cancer markers were most likely to generate an immune response. They fell into two major categories: the way the neoantigen is presented on the cancer cell and how the neoantigen is recognized by the immune system.

When the data model emphasizing these five characteristics was put to the test against another set of cancer samples, it accurately predicted 75 percent of effective neoantigen targets and filtered out 98 percent of ineffective ones. When the model was reapplied to participating teams’ algorithms, the predictions measurably improved.

The full TESLA dataset, the largest of its kind, is available freely to the research community in order to accelerate personalized therapy development and even improve efficacy of these treatments for cancer patients worldwide.