An AI-powered pipeline for personalized cancer vaccines

Ludwig Cancer Research scientists have developed a full, start-to-finish computational pipeline that integrates multiple molecular and genetic analyses of tumors and the specific molecular targets of T cells and harnesses artificial intelligence algorithms to use its output to design personalized cancer vaccines for patients.

The design, validation and comparative assessment of this computational suite, NeoDisc, are detailed in the current issue of Nature Biotechnology in a publication led by Florian Huber and Michal Bassani-Sternberg of the Lausanne Branch of the Ludwig Institute for Cancer Research.

“NeoDisc provides unique insights into the immunobiology of tumors and the mechanisms by which they evade targeting by cytotoxic T cells of the immune system,” said Bassani-Sternberg.

“These insights are invaluable to the design of personalized immunotherapies, and the analytical and computational pipeline at the heart of NeoDisc is already being used here in Lausanne for clinical trials of personalized cancer vaccines and adoptive cell therapies.”

Many cancer types harbor multiple random mutations that should make them more visible to the immune system. Such mutations generate aberrant proteins that cells, even cancerous ones, are programmed to cut into short pieces—known as peptides—and “present” as antigens to invite an attack by patrolling T cells.

The great diversity of these “neoantigens” is one of the reasons why patients respond so variably to immunotherapies. On the other hand, neoantigens can be harnessed to develop vaccines and other types of immunotherapies tailored to uniquely target each patient’s tumors. Personalized treatments of this kind are now being developed by researchers around the world.

Such efforts are technically challenging because not all neoantigens are recognized by a given patient’s T cells, and even many that are recognized fail to elicit a sufficiently potent T cell attack. One approach to designing personalized vaccines and cell therapies thus involves the identification of neoantigens most likely to provoke a vigorous T cell assault.

This requires sophisticated, large-scale analyses of mutations that generate potential neoantigens, the molecular scaffolding (known as HLA molecules) that presents them to T cells and the molecular characteristics that enable recognition by T cell receptors. Bassani-Sternberg is among the pioneers of this field, a high-tech marriage of large-scale biochemical and computational analysis known as “immunopeptidomics.”

The design of personalized immunotherapies is also aided by genomic analysis of both the tumor and blood cells that represent the healthy genome of the patient, the large-scale analysis of gene expression known as “transcriptomics” as well as the sensitive analysis of the so-called immunopeptidome with mass spectrometry.

Until now, however, these powerful technologies have never been integrated in a single computational pipeline to predict which neoantigens identified in a patient’s tumors should be employed as vaccines or otherwise harnessed for personalized immunotherapies.

Beyond that, neoantigens are not the only type of antigens available for immunotherapeutic targeting. Cancer cells also erroneously express as proteins bits of ordinarily noncoding DNA, genes normally expressed only during development, other aberrantly expressed gene products and viral antigens, in cases of virally induced tumors—all of which can provoke immune attack.

“NeoDisc can detect all these distinct types of tumor-specific antigens along with neoantigens, apply machine learning and rule-based algorithms to prioritize those most likely to elicit a T cell response, and then use that information to design a personalized cancer vaccine for the relevant patient,” said Huber.

NeoDisc additionally ranks the potential antigens it detects and generates visualizations of cancer cell heterogeneity within tumors.

“Notably, NeoDisc can also detect potential defects in the machinery of antigen presentation, alerting vaccine designers and clinicians to a key mechanism of immune evasion in tumors that can compromise the efficacy of immunotherapy,” said Bassani-Sternberg. “This can help them select patients for clinical studies who are likely to benefit from personalized immunotherapy, a capability that is also of great importance to optimizing patient care.”

The researchers additionally show in their study that NeoDisc provides a more accurate selection of effective cancer antigens for vaccines and adoptive cell therapies than do other computational tools currently used for that purpose.

To further enhance NeoDisc’s accuracy, the researchers will continue feeding it data obtained from a variety of tumors and integrate additional machine-learning algorithms to the software suite to advance its training and improve its predictive accuracy.