IMMUNOTAR - Integrative prioritization of cell surface targets for cancer immunotherapy

Abstract

Cancer remains a leading cause of mortality globally. Recent improvements in survival have been facilitated by the development of less toxic immunotherapies; however, identifying targets for immunotherapies remains a challenge in the field. To address this challenge, we developed IMMUNOTAR, a computational tool that systematically prioritizes and identifies candidate immunotherapeutic targets. IMMUNOTAR integrates user-provided RNA-sequencing or proteomics data with quantitative features extracted from publicly available databases based on predefined optimal immunotherapeutic target criteria and quantitatively prioritizes potential surface protein targets. We demonstrate the utility and flexibility of IMMUNOTAR using three distinct datasets, validating its effectiveness in identifying both known and new potential immunotherapeutic targets within the analyzed cancer phenotypes. Overall, IMMUNOTAR enables the compilation of data from multiple sources into a unified platform, allowing users to simultaneously evaluate surface proteins across diverse criteria. By streamlining target identification, IMMUNOTAR empowers researchers to efficiently allocate resources and accelerate immunotherapy development.

Keywords: RNA-sequencing; bioinformatics; cancer; immunotherapy; pediatric cancer; proteomics.

Figure 1:. IMMUNOTAR feature extraction and surface protein scoring scheme.

A) The pre-defined criteria for an ideal immunotherapeutic target. B) Summary of features extracted from the user-input expression dataset and the publicly available databases incorporated in IMMUNOTAR. C) The steps of analysis within IMMUNOTAR. The first step involves extracting quantitative features from the user input and the integrated databases to form a scoring data matrix. In the second step, IMMUNOTAR can rescale and handle missing values of the extracted features using the methods listed. After rescaling and handling missing values of the features, the user has the option of applying a curving function to the feature values. After curving the feature values, a weight parameter can be applied and the final score of each gene is calculated as the sum of the weighted-curved-features.

Figure 2:. IMMUNOTAR evaluation, optimization, and validation datasets.

A) The mean average precision (MAP) score is used to evaluate the performance of IMMUNOTAR. The MAP score depends on the known-positive and known-negative targets assigned to each phenotype and is calculated based on the ranking of those targets after running IMMUNOTAR. B) Optimization methods listed can be used to change the weights and curves of features. The goal of the optimization is to rank the known-positive targets higher in the results thus increasing the overall MAP score of the algorithm. C) Optimized parameters are generated using a multi-cancer proteomic dataset and applied to other datasets to validate the approach. Validation datasets include Ewing sarcoma surface proteomics, multiple myeloma surface proteomics, and neuroblastoma full proteome.

Figure 3:. IMMUNOTAR optimization using pediatric pan-cancer expression data.

A) The pediatric cancer phenotypes included in the optimization scheme, listing the number of cell lines and the number of known-positive surface protein targets extracted from the drug databases TTD and ADCdb (Supplemental Table S2). B) Optimization parameters were generated using two approaches, the first being a multi-cancer optimization approach and the second being a phenotype-specific optimization approach. C) Using the multi-cancer optimization parameters, the MAP score for each individual phenotype increased at varying levels within phenotypes. The phenotype-specific optimization approach increased the MAP score further for each phenotype. D) While the phenotype-specific parameters performed better on the phenotype itself, they did not perform as well on other phenotypes. Overall, the multi-cancer optimization parameters performed best across all phenotypes with an average MAP score of 14%.

Figure 4:. Validation of IMMUNOTAR using EwS surface proteomics data.

A) IMMUNOTAR results from the EwS surface proteomics data using the multi-cancer optimization parameter values. The labeled proteins include targets that are known within EwS and highlighted by Mooney and colleagues. The blue labeled targets are novel targets discovered by Mooney and colleagues. The purple dashed line represents the cut-off for the top 5% of proteins scored by IMMUNOTAR. B) Heatmap showing the top 5% scored IMMUNOTAR targets after implementing a restricted normal-tissue expression filter within IMMUNOTAR. We marked targets that were prioritized by Mooney and colleagues scoring scheme and targets that were fed as known-positives in the IMMUNOTAR analysis. C) The GSEA enrichment scores comparing three sets of prioritized targets in EwS. The IMMUNOTAR scores of each target were compared to the target scores from the Mooney et al. scoring scheme. IMMUNOTAR outputs using both the multi-cancer optimization parameter values and the phenotype-specific optimization parameter values were compared to the Mooney et al. scoring scheme, separately. *p-value <= 0.05. D) Protein quantification of CADM1 in EwS surface proteomics dataset showing comparable abundance to other EwS-specific known-positive targets.

Figure 5:. Validation of IMMUNOTAR using MM surface-proteomics data.

A) IMMUNOTAR results from the MM surface proteomics dataset using the multi-cancer optimization parameters. Labeled proteins are ones that were ranked in the top-three scoring brackets by Ferguson and colleagues. The purple dashed line represents the top 5% scoring targets. The blue labeled target was prioritized and validated by Ferguson et al. as a novel target for MM. The red labeled target, TXNDC11 is a prioritized protein that was a false-positive proved by functional validation. B) The top 5% scoring targets using IMMUNOTAR, marking targets prioritized in IMMUNOTAR and the top-three scoring brackets and targets that are known-positive targets per TTD and ADCdb. C) The GSEA enrichment scores comparing three sets of prioritized targets in MM. The IMMUNOTAR scores of each target were compared to the target scores from the Ferguson et al. scoring scheme. *p-value <= 0.05.

Figure 6:. IMMUNOTAR prioritized targets in NBL using full proteome data and phenotype-specific optimization parameters.

A) Evaluating the top 5% scoring targets per IMMUNOTAR analysis after applying phenotype-specific parameters and the restrictive normal-tissue expression filter. Highlighting known, top-scoring target, ALK. B) Protein quantification of KCNH1 in full proteome dataset showing comparable abundance to other NBL-specific known-targets. C) Normal tissue expression per GTEx for KCNH1 reveals limited normal tissue expression, with some expression in healthy brain. Tissues with median expression ≥ 5 TPM are labeled.