AI Empowers Oncology: Breakthrough Antigen Discoveries Herald a New Era in Cancer Therapy

With the advent of immune checkpoint inhibitors and CAR-T therapy, significant progress has been made in cancer treatment. Merck's pembrolizumab, an anti-PD-1 therapy, is expected to lead as the pharmaceutical champion of 2023. Despite challenges, the CAR-T therapy market is rapidly expanding. The success of these treatments marks the dawn of immunotherapy in cancer care.

To further advance immunotherapy, the medical community tirelessly seeks more treatment targets. Post-mRNA splicing products, generating a plethora of proteins or peptides, hold potential as novel targets for cancer treatment. Splicing neoantigens are an unexplored treasure trove for scientists.

While various tools have emerged for identifying splicing neoantigens, they still have usability and accessibility limitations. A recent breakthrough from Cincinnati Children's Hospital, the Splicing Neo Antigen Finder (SNAF), promises to accelerate research in this field with its user-friendly interface and open-source nature.

AI Once Again Demonstrates Its Power

In recent years, AI technology has excelled in biomedical research, notably with Alphafold's emergence driving significant strides in structural biology. SNAF, a product of AI, is another milestone. Leveraging Bayesian models and deep learning, researchers crafted two tools: BayesTS for tumor specificity prediction and DeepImmuno for immunogenicity prediction. These tools birthed two splicing neoantigen discovery pipelines, SNAF-T and SNAF-B, tailored for T-cell and B-cell therapy, respectively.

To validate SNAF's functionality, researchers scrutinized publicly available cancer patient databases using it. In ovarian cancer and melanoma databases, they unearthed numerous potential neoantigens. Subsequent mass spectrometry validation of 36 peptides identified 27 successfully. Additionally, seven neoantigens with top algorithm scores emerged.

These seven potential neoantigens stem from diverse splicing mechanisms. Notably, HAAASFETL, a splicing neoantigen from the FCRLA gene, arises from exon skipping splicing, boasting a meager BayesTS score of 0.03 in non-tumor tissues. Despite FCRLA's low expression in tumors, spliced HAAASFETL is prevalent in over one-third of melanoma patients, underscoring its tumor specificity. The remaining six neoantigens, originating from different genes, also underwent mass spectrometry validation, affirming SNAF's predictive accuracy.

The Clinical Significance of Splicing Neoantigens

Given SNAF's precise identification of splicing neoantigens, understanding their clinical significance is imperative. To elucidate this, researchers expanded their analysis to 500 melanoma samples, pinpointing an average of 528 tumor-specific splicing events and predicting 1090 MHC-binding peptides. DeepImmuno further refined this, reducing noise by 16% to 915 peptides.

Subsequently, researchers compared the clinical outcomes of high MHC-binding neoantigen groups to low MHC-binding groups, revealing significantly poorer survival outcomes in the former. This trend extended to patients with high neoepitope splicing and immunogenic peptide burdens. Gene expression analysis identified 597 upregulated and 227 downregulated genes in high splicing neoantigen burden groups, many linked to tumor occurrence, DNA repair, and therapy resistance. Thus, high splicing neoantigen burden groups may derive less benefit from conventional cancer therapies.

Further analysis uncovered 2,970 splice variants associated with adverse prognosis, including 108 present in over 15% of patients, dubbed "shared neoantigens." These can serve as biomarkers aiding treatment assessment and guiding future directions. However, differences in biomarker functionality across databases exist. Some neoantigens show negative correlations with survival rates in TCGA data, yet positive correlations in Van Allen database analyses. Nonetheless, the data underscore the immense clinical value of SNAF-identified neoantigens.

With SNAF's value confirmed, researchers probed upstream factors contributing to varying neoantigen burdens among patients. Analyzing RNA-binding protein knockdown databases, they identified 209 proteins with reduced activity in high neoantigen burden groups. Despite upregulated expression levels in these groups, these proteins failed to effectively splice mRNA, resulting in heightened intron retention, a potential driver of neoantigen burden discrepancies.

The Therapeutic Value of Neoantigens

Having set the stage, the true value of SNAF lies in its validation through T-cell and B-cell therapy. To develop T-cell therapy targeting neoantigens, the crucial proof required is demonstrating that SNAF-T discovered antigens can trigger T-cell responses by being presented by MHC.

Researchers began by scrutinizing 940 T-cell "shared neoantigens," finding their detection rate significantly higher than individually unique ones. Over 34% of neoantigens were backed by immunopeptidomic data, with 98 confirmed in analyses of at least 15% of patients. Peptide sequence classification revealed patterns, like shared neoantigen sequences often ending with lysine and phenylalanine, aiding future identification.

Next, five neoantigens were chosen for cellular experiments, all provoking specific T-cell responses in vitro with greater accuracy than past prediction methods.

To advance SNAF-based therapies, neoantigens must originate from tumor cells, not the tumor microenvironment. Single-cell sequencing confirmed that most shared neoantigens came from tumor cells.

Lastly, researchers explored predicting extracellular neoepitopes with SNAF-B and Alphafold2, identifying splice variants containing these epitopes, termed ExNeoEpitopes. Fluorescent labeling and expression of full-length sequences confirmed ExNeoEpitopes on cell membranes, opening new avenues for monoclonal antibody and CAR-T cell therapies.

As an open-source tool, SNAF augments drug discovery, with AI-driven drugs expected to transition from labs to clinics soon, benefiting countless cancer patients.