485 Background: The role of poly-ADP-ribose polymerase (PARP) inhibitors is currently under investigation as a potential therapeutic option for AESCC. Homologous recombination deficiency signature status (HRDSig) positivity has been shown to predict biallelic loss of BRCA1/2 which may simultaneously portend a worse prognosis and a sensitivity to poly-ADP-ribose polymerase (PARP) inhibitors. In this study, we attempt to characterize the genomic alterations present in AESCC based on HRDSig status utilizing comprehensive genomic profiling (CGP) techniques. Methods: 2,029 cases of AESCC underwent comprehensive genomic profiling with an examination of all genomic alterations (GA). MSI high status, tumor mutation burden (TMB) levels, genomic ancestry and genomic trinucleotide signatures were determined from the sequencing data. HRDSig status was calculated using a broad set of genome-wide copy number features (PMID 37769224). Results were compared using the Fisher exact system with the Benjamini-Hochberg adjustment to correct for false discovery. Results: 154 (7.6%) of the 2029 AESCC cases featured a positive HRDSig status (HRDSig+). The age (66-67) and gender (58%-63% male) distribution were similar in the HRDSig+ and HRDSig- AECSS cases as was the frequency of GA/tumor (9 for both). The median TMB was higher in the HRDSig+ (6.3 vs 3.8; P<.0001) as was frequency of TMB > 10 mutations/Mb (19.5% vs 10.0%; P=.004). An APOBEC trinucleotide signature was also more frequent in the HRDSig+ AESCC (11.0% vs 5.4%; P=.05). As anticipated, GA in genes associated with HRD including BRCA1 (10.4% vs 1.4%; P<.0001) and BRCA2 (14.3% vs 2.0%; P<.0001) was present. MTOR pathway activating mutations were also more frequent in the HRDSig+ AESCC group including PTEN (13.0% vs 7.3%; P=.06). Conclusions: With a 7.6% frequency, HRDSig+ status is a relatively rare event in AESCC. CGP with determination of HRDSig status in AESCC may prove useful in tailoring PARP inhibitor-based treatment regimens and may potentially uncover other genomic alterations that can aid in designing targeted therapy combinations in future. Characteristics of comprehensive genomic profiling of clinically advanced esophageal squamous cell carcinoma by HRDSig status. AESCC HRDSig- (n=1875) AESCC HRDSig+ (n=154) P-value Pathogenic genomic alterations BRCA1 1.4% (26) 10.4% (16) 0.00 BRCA2 2.0% (36) 14.3% (22) 0.00 PTEN 7.3% (137) 13.0% (20) 0.06 COSMIC trinucleotide signature APOBEC 5.4% (101) 11.0% (17) 0.05 Tumor mutational burden (TMB) Median TMB (range) (IQR) 3.8 (0-85) (2.5-6.3) 6.3 (0-61) (3.8-8.8) 0.00 TMB≥10 mut/Mb 10.0% (188) 19.5% (30) 0.00

239 Background: Inactivating genomic alteration (GA) of the TP53 tumor suppressor gene leads to inactivation of p53 protein and is currently the most frequently associated GA across all cancers. Colorectal cancer is a common and lethal subtype of cancer, in the top 5 for incidence, prevalence and cancer-related deaths worldwide. TP53 is the second most frequent GA identified in clinically advanced CRC. It has been postulated that GA in TP53 not only affects the molecular biology of malignant cells but also impacts the tumor microenvironment in CRC. This GA is currently untargetable by approved drugs however, recently a novel approach designed to revert to wild type functioning of TP53 inactivated by Y220C base substitution has gained significant clinical interest. Methods: 67,301 cases of clinically advanced CRC underwent hybrid capture based comprehensive genomic profiling to assess all classes of GA. Cases were sequenced to a mean coverage depth of 650X with microsatellite instability (MSI) status and tumor mutation burden (TMB) determined from the sequencing data and PD-L1 expression measured by immunohistochemistry using the DAKO tumor proportional staining system (low positive set at 1-49% staining and high positive set at ≥50% staining). Results: 422 (0.6%) of the CRC featured the TP53 Y220C GA (Y220C+). When compared with the CRC that were TP53 Y220C negative (Y220C-), the Y220C+ cases were of similar age, gender and number of GA per tumor. The Y220C- cases has significantly higher frequencies of MSI status (6.1% vs 2.2%; p=.0007) and TMB levels greater >10 mutations/Mb (8.7% vs 4.7%; p=.004). The Y220C+ featured lower frequencies of GA in PIK3CA (12.9% vs 18.6%; p=.002), BRAF (8.8% vs 10.1%; NS) and PTEN (7.4% vs 8.3%, NS) and a higher frequency of ERBB2 amplification (5.0% vs 2.7%; p=.004). GA in APC , KRAS , NRAS , BRCA2 , EGFR and MET were similar in both Y220C+ and Y220C- CRC. PD-L1 low expression was similarly infrequent (12.1% to 16.7% range) in both groups. Conclusions: The potentially targetable TP53 Y220C short variant mutation is uncommon in clinically advanced CRC but is associated with unique genomic and biomarker characteristics. Further research might elucidate subpopulations or CRC subtypes with a greater frequency of this mutation who might benefit from a novel targeted therapy. CRC TP53 Y220C+(422 Cases) CRC TP53 Y220C-(66,879 Cases) P Value Gender 41.2% F/ 58.8% M 44.4% F/55.6% M NS Median Age (range) 62 (23-89+) 61 (8-89+) NS GA/tumor 6.1 6.5 NS TP53 non-Y220C GA 16.1% 75.7% <.0001 PIK3CA 12.9% 18.6% .002 ERBB2 6.2% (5.0% amp) 5.1% (2.7% amp) .004 (amp) MSI High 2.2% 6.1% .0007 TMB > 10 mut/Mb 4.7% 8.7% .004 PD-L1 Low (1-49%TPS)/ PD-L1 High (>50% TPS) 16.7% (132 cases)/2.3% 12.1% (20,759 cases)/1.4% NS

Immune checkpoint inhibitors (ICI) have become integral to treatment of non-small cell lung cancer (NSCLC). However, reliable biomarkers predictive of immunotherapy efficacy are limited. Here, we introduce HistoTME, a novel weakly supervised deep learning approach to infer the tumor microenvironment (TME) composition directly from histopathology images of NSCLC patients. We show that HistoTME accurately predicts the expression of 30 distinct cell type-specific molecular signatures directly from whole slide images, achieving an average Pearson correlation of 0.5 with the ground truth on independent tumor cohorts. Furthermore, we find that HistoTME-predicted microenvironment signatures and their underlying interactions improve prognostication of lung cancer patients receiving immunotherapy, achieving an AUROC of 0.75[95% CI: 0.61-0.88] for predicting treatment responses following first-line ICI treatment, utilizing an external clinical cohort of 652 patients. Collectively, HistoTME presents an effective approach for interrogating the TME and predicting ICI response, complementing PD-L1 expression, and bringing us closer to personalized immuno-oncology.

480 Background: Despite recent advances in treatment of EGAC, prognosis continues to be poor with reported 5-year survival of less than 30%. EGAC is more frequently diagnosed at an old age with the median age at diagnosis around 65 years. However, approximately 10% of EGAC are diagnosed at a younger age of less than 50 years and those patients present with more advanced stages resulting in worse prognosis. Our study investigated the genomic landscape of EGAC diagnosed under the age of 30 years to better understand the characteristics of EGAC occurring in very young population. Methods: 8,001 cases of clinically advanced EGAC who underwent comprehensive genomic profiling (CGP) from 2012 to 2024 were included in the study. Patients aged less than 30 years were classified as ‘EGAC-young’ and patients aged 50 and older were classified as ‘EGAC-old’. All classes of genomic alterations (GA), microsatellite instability high (MSI-high) status, tumor mutation burden (TMB), genomic ancestry, genomic signature, and homologous recombination defect signature (HRDSig) were determined from the sequencing data. PD-L1 was measured by IHC using the Dako 22C3 tumor proportional score (TPS) system. Comparisons utilized the Fisher Exact method with the Benjamini-Hochberg adjustment to reduce the false discovery rate. Results: Out of 8,001 cases, 7389 (92.4%) were classified as ‘EGAC-old’ and 26 (0.3%) as ‘EGAC-young’. Both groups had a higher frequency of male patients but there was no significant difference between the two groups (86.4% vs 76.9%; NS). Median GA per tumor was 6 in both groups. The ‘EGAC-old’ featured a significantly higher frequency of EUR ancestry (92.0% vs 73.1%; P=0.026). There were no significant differences in clinically important GA between both groups, including similar ERBB2 GA (21.2% vs 30.8%; NS) and FGFR3 GA (1.2% vs 7.7%; NS). There was a trend of more frequent GA of AKT2, CCNE1, and KEAP1 noted in ‘EGAC-young’ (0.9% vs 11.5%, 8.5% vs 26.9%, 1.3% vs 11.5%, respectively; P=0.071). MSI-high status was 3.1% in the ‘EGAC-old’ but not detected in the ‘EGAC-young’ (NS). The median TMB was higher in the ‘EGAC-old’ vs ‘EGAC young’ (3.8 mutations/Mb vs 1.9 mutations/Mb; P=.003). The frequencies of a positive HRDSig were similar in both groups (5.6% vs 8.0%; NS). There were no significant differences in COSMIC trinucleotide signatures. PD-L1 expression was identified in greater than 20% of the ‘EGAC-old’ but was not measured in the ‘EGAC-young’. Conclusions: Clinically advanced EGAC in young patients under the age of 30 years features a genomic landscape that differs from EGAC identified in older patients, including lower TMB and a trend of increased GA frequency of genes related to poor prognosis in other types of cancer. Further studies are warranted to utilize CGP findings in identifying potential implications for treatment and prognostication in these rare cases of EGAC occurring in very young patients.

Successful cancer resection is limited by the inability to differentiate between cancer and normal tissue intraoperatively. Desorption electrospray ionization mass spectrometry imaging (DESI-MSI) is an emerging and powerful analytical technique that offers a rapid and low cost approach for assessing surgical margins by generating detailed metabolic profiles. However, exploiting this data for tissue characterization based on molecular signals requires machine learning methods to handle its complexity. In this work, we utilize machine learning models for the characterization of tissue using DESI-MSI data obtained from prostate tissue samples. We use ViPRE, a novel open-source software, to annotate a large DESI-MSI dataset. We explore various machine learning models and train test schemes for cancer classification. Cross-validation of our models result in high balanced accuracy, sensitivity and specificity for cancer classification. Furthermore, we simulate the prospective application of perioperative tissue characterization, generating a qualitative visual prediction for whole slides that match pathology annotations. Finally, the application of linear transformation and classification algorithms on DESI-MSI data effectively distinguished between the molecular profiles associated with different cancer grades. Our findings highlight the promise of combining machine learning with large DESI-MSI datasets for tissue characterization, thereby improving surgical margin precision.

Background: The effectiveness of the clinical outcome of CN (Cytoreductive Nephrectomy) in cases of mccRCC (Metastatic Clear Cell Renal cell Carcinoma) is still uncertain despite two trials, SURTIME and CARMENA. These trials, conducted with Sunitinib as the standard treatment, did not provide evidence supporting the use of CN. Methods: We queried the NCDB for stage IV mccRCC patients between the years of 2004 to 2020, who received (immunotherapy) IO with or without nephrectomy. Overall survival (OS) was calculated among three groups of IO alone, IO followed by CN (IOCN), CN followed by IO (CNIO). Cox models compared OS by treatment group after adjusting for sociodemographic, health, and facility variables. Results: From 1,549,101 renal cancer cases, 7983 clear and nonclear cell renal cell carcinoma cases were identified. After adjusting for sociodemographic and health covariates, patients who received IO followed by CN or CN followed by IO had a respective 64% (adjusted Hazard Ratio [aHR] = 0.36, 95% CI = 0.300.43, P = .006] and 47% (aHR = 0.53, 95% CI = 0.49-0.56, P = .001) mortality risk reduction respectively compared to patients who received IO alone. Compared to White adults, individuals who identified as Black exhibited 17% higher risk mortality (aHR = 1.17, 95% CI = 1.06-1.30, P = .002). Patients who received CN prior to IO had a 59% associated mortality risk compared to patients who received IO followed by CN who had a lower risk, 35.7% ( P < .001). Conclusions: Patients receiving CN regardless of sequence with IO did better than IO alone in this national registry-based adjusted analysis for mccRCC. Presently available data indicates that the combination of CN and IO holds promise for enhancing clinical results in patients with mRCC.

IntroductionThe 2023 Coffey-Holden Prostate Cancer Academy (CHPCA) Meeting, themed "Disrupting Prostate Cancer Research: Challenge Accepted," was convened at the University of California, Los Angeles, Luskin Conference Center, in Los Angeles, CA, from June 22 to 25, 2023.MethodsThe 2023 marked the 10th Annual CHPCA Meeting, a discussion-oriented scientific think-tank conference convened annually by the Prostate Cancer Foundation, which centers on innovative and emerging research topics deemed pivotal for advancing critical unmet needs in prostate cancer research and clinical care. The 2023 CHPCA Meeting was attended by 81 academic investigators and included 40 talks across 8 sessions.ResultsThe central topic areas covered at the meeting included: targeting transcription factor neo-enhancesomes in cancer, AR as a pro-differentiation and oncogenic transcription factor, why few are cured with androgen deprivation therapy and how to change dogma to cure metastatic prostate cancer without castration, reducing prostate cancer morbidity and mortality with genetics, opportunities for radiation to enhance therapeutic benefit in oligometastatic prostate cancer, novel immunotherapeutic approaches, and the new era of artificial intelligence-driven precision medicine.DiscussionThis article provides an overview of the scientific presentations delivered at the 2023 CHPCA Meeting, such that this knowledge can help in facilitating the advancement of prostate cancer research worldwide.

11131 Background: The impact of Eastern Cooperative Oncology Group Performance Score (ECOG) on immunotherapy (IO) outcomes is intricate, as historically clinical trials have primarily enrolled patients with ECOG scores of 0 or 1. The National Comprehensive Cancer Network (NCCN) advises against administering IO to patients with ECOG scores of 2 or higher. However, conflicting findings from research studies and variations in real-world clinical practice complicate this relationship. Our study aims to elucidate the association between ECOG status and the outcomes of immunotherapy. Methods: Data collected from SUNY Upstate Medical University treated cancer patients who underwent immunotherapy and chemotherapy was used for analysis. The Kaplan-Meier method and Cox regression were used to analyze survival probability based on therapy, age, and ECOG. Results: Of 813 patients included in the study, 46.99% (n=382) received immunotherapy. 51.85% (n=421) were females. The mean age was 64.1 years in IO group and 67.7 years in non-IO/chemo group. 52.5% (n=187) of patients who were ≤ 64 years received IO compared to only 42.7% (n=195) of patients who were ≥ 65 years (p=0.005). 48.6% (n=118) of patients with ECOG score 0 received IO, while only 40.5% (n=72) patients with ECOG score of ≥3 received IO (p=0.279). Patients who received IO with ECOG score of ≥3 had higher probability of survival compared to other groups (p<0.0001). While, in non-IO group, patients with ECOG 0 had higher probability of survival compared to ECOG 3 (HR 1.773, 1.209 – 2.600). There was no difference in overall survival in patients receiving IO based on age categories (p=0.2627, HR 1.148, 0.901 – 1.462). Conclusions: Our findings suggest that relying solely on ECOG status to determine eligibility for immunotherapy may be overly restrictive. Patients with significant comorbidities could still derive benefits from immunotherapy. Further investigation is warranted to comprehensively assess the influence of ECOG status on immunotherapy outcomes.

8530 Background: Immune checkpoint inhibitors (ICIs) targeting Programmed Death 1 (PD1) or Programmed Death Ligand 1 (PDL1) have revolutionized non-small cell lung cancer (NSCLC) treatment, yet only 30% of patients respond effectively. Although approved for patient selection, using PDL1 expression as a biomarker of treatment response is controversial in predicting clinical outcomes, highlighting the need for more robust alternatives. Methods: We explore the potential of AI-based approaches using H&E-stained digitized images of resection and biopsy samples to enhance the prediction of ICI responses in NSCLC. A dataset of 692 NSCLC patients (n=1325 slides) treated with anti-PD1/PDL1 was collected from SUNY Upstate Medical University, focusing on a subset of 166 patients receiving first-line anti-PD1/PDL1 treatment for AI model training and evaluation. An attention-based deep neural network was trained to predict treatment responses of patients from whole slide H&E images while being challenged with additional auxiliary tasks such as interferon gamma gene expression prediction in order to enable a comprehensive analysis of the tumor microenvironment (TME) and mitigate overfitting. Overall, 60% of the SUNY cohort (N=104; 26 Partial/Complete response, 78 stable/Progressive disease) and the publicly available TCGA-NSCLC cohort (N=986) were-utilized for multi-task training. The remaining 40% of the SUNY cohort (N=62; 16 Partial/Complete response, 46 stable/Progressive disease) was used for testing. Training was done in 5-fold cross-validation. The final model was chosen as an ensemble of models trained over all 5 folds. Train and test splits were generated using a stratified random sampling approach, ensuring that the proportion of responders versus non-responders remained unchanged across splits. Performance was evaluated utilizing percentage accuracy (all true positive values), precision (positive predictive value), recall (sensitivity) and F1 scores (mean value of precision and recall). Attention maps generated by the AI model were used to highlight relevant spatial regions of the TME significantly associated with ICI responses. Results: Our multi-task attention-based approach achieves an overall predictive accuracy of 79%, precision of 0.53, recall of 0.24, and an F1 score of 0.33 on the held-out test set. In contrast, pathologist derived PDL1 IHC scores ≥50% exhibit an accuracy of 47%, precision of 0.09, recall of 0.125, and an F1 score of 0.11. Conclusions: Initial findings indicate that attention-based multitask learning of NSCLC H&E- images could uncover crucial tumor intrinsic and microenvironmental features that are predictive of ICI responses in patients. Inclusion of additional clinical and molecular data for training and validation holds potential to further improve ICI response prediction accuracy using AI-derived biomarkers and classifiers.

Purpose: Latent grade group ≥2 prostate cancer can impact the performance of active surveillance protocols. To date, molecular biomarkers for active surveillance have relied solely on RNA or protein. We trained and independently validated multimodal (mRNA abundance, DNA methylation, and/or DNA copy number) biomarkers that more accurately separate grade group 1 from grade group ≥2 cancers. Materials and Methods: Low- and intermediate-risk prostate cancer patients were assigned to training (n=333) and validation (n=202) cohorts. We profiled the abundance of 342 mRNAs, 100 DNA copy number alteration loci, and 14 hypermethylation sites at 2 locations per tumor. Using the training cohort with cross-validation, we evaluated methods for training classifiers of pathological grade group ≥2 in centrally reviewed radical prostatectomies. We trained 2 distinct classifiers, PRONTO-e and PRONTO-m, and validated them in an independent radical prostatectomy cohort. Results: PRONTO-e comprises 353 mRNA and copy number alteration features. PRONTO-m includes 94 clinical, mRNAs, copy number alterations, and methylation features at 14 and 12 loci, respectively. In independent validation, PRONTO-e and PRONTO-m predicted grade group ≥2 with respective true-positive rates of 0.81 and 0.76, and false-positive rates of 0.43 and 0.26. Both classifiers were resistant to sampling error and identified more upgrading cases than a well-validated presurgical risk calculator, CAPRA (Cancer of the Prostate Risk Assessment; P < .001). Conclusions: Two grade group classifiers with superior accuracy were developed by incorporating RNA and DNA features and validated in an independent cohort. Upon further validation in biopsy samples, classifiers with these performance characteristics could refine selection of men for active surveillance, extending their treatment-free survival and intervals between surveillance.

Mass Spectrometry Imaging (MSI) is a powerful tool capable of visualizing molecular patterns to identify disease markers in tissue analysis. However, data analysis is computationally heavy and currently time-consuming as there is no single platform capable of performing the entire preprocessing, visualization, and analysis pipeline end-to-end. Using different software tools and file formats required for such tools also makes the process prone to error. The purpose of this work is to develop a free, open-source software implementation called “Visualization, Preprocessing, and Registration Environment” (ViPRE), capable of end-to-end analysis of MSI data. ViPRE was developed to provide various functionalities required for MSI analysis including data import, data visualization, data registration, Region of Interest (ROI) selection, spectral data alignment and data analysis. The software implementation is offered as an open-source module in 3D Slicer, a medical imaging platform. It is also designed for flexibility and usability throughout the user experience. ViPRE was tested using sample MSI data to evaluate the computational pipeline, with the results showing successful implementation of its functionalities and end-to-end usage. A preliminary usability test was also performed to assess user experience, with findings showing positive results. ViPRE aspires to satisfy the need for a single-stop comprehensive interface for MSI data analysis. The source code and documentation will be made publicly available.

Purpose: Latent Grade Group (GG) ≥2 prostate cancer can impact the performance of active surveillance (AS) protocols. To date, molecular biomarkers for AS have relied solely on RNA or protein. We trained and independently validated multimodal (mRNA abundance, DNA methylation, and DNA copy number) biomarkers that more accurately separate GG1 from GG≥2 cancers. Materials and Methods: Low- and intermediate-risk prostate cancer patients were assigned to training (n=333) and validation (n=202) cohorts. We profiled the abundance of 342 mRNAs, 100 DNA copy number aberration (CNA) loci and 14 hypermethylation sites at two locations per tumor. Using the training cohort with cross- validation, we evaluated methods for training classifiers of pathologic GG≥2 in centrally reviewed radical prostectomies (RPs). We trained two distinct classifiers, PRONTO-e and PRONTO-m, and validated them in an independent RP cohort. Results: PRONTO-e comprises 353 mRNA and CNA features. PRONTO-m includes 94 clinical, mRNAs, CNAs and methylation features at 14 and 12 loci, respectively. In independent validation, PRONTO-e and PRONTO-m predicted GG≥2 with respective true positive rates of 0.81 and 0.76, false positive rates of 0.43 and 0.26. Both classifiers were resistant to sampling error and identified more upgraded men than a well-validated pre-surgical risk calculator, CAPRA (p <0.001). Conclusions: Two GG classifiers with superior accuracy were developed by incorporating RNA and DNA features and validated in an independent cohort. Upon further validation in biopsy samples, classifiers with these performance characteristics could refine selection of men for AS, extending their treatment-free survival and intervals between surveillance. Citation Format: Anna Y. Lee, David M. Berman, Robert Lesurf, Palak G. Patel, Walead Ebrahimizadeh, Jane Bayani, Laura A. Lee, Nadia Boufaied, Shamini Selvarajah, Tamara Jamaspishvili, Karl-Philippe Guérard, Dan Dion, Atsunari Kawashima, Gina M. Clarke, Nathan How, Chelsea L. Jackson, Eleonora Scarlata, Khurram Siddiqui, John B.A. Okello, Armen G. Aprikian, Madeleine Moussa, Antonio Finelli, Joseph Chin, Fadi Brimo, Glenn Bauman, Andrew Loblaw, Vasundara Venkateswaran, Ralph Buttyan, Simone Chevalier, Axel Thomson, Paul C. Park, D. Robert Siemens, Jacques Lapointe, Paul C. Boutros, John M.S. Bartlett. Multimodal biomarkers that predict the presence of Gleason pattern 4: Potential impact for active surveillance [abstract]. In: Proceedings of the AACR Special Conference: Advances in Prostate Cancer Research; 2023 Mar 15-18; Denver, Colorado. Philadelphia (PA): AACR; Cancer Res 2023;83(11 Suppl):Abstract nr B046.

Phosphatase and tensin homolog (PTEN) loss is associated with adverse outcomes in prostate cancer and can be measured via immunohistochemistry. The purpose of the study was to establish the clinical application of an in-house developed artificial intelligence (AI) image analysis workflow for automated detection of PTEN loss on digital images for identifying patients at risk of early recurrence and metastasis. Postsurgical tissue microarray sections from the Canary Foundation (n = 1264) stained with anti-PTEN antibody were evaluated independently by pathologist conventional visual scoring (cPTEN) and an automated AI-based image analysis pipeline (AI-PTEN). The relationship of PTEN evaluation methods with cancer recurrence and metastasis was analyzed using multivariable Cox proportional hazard and decision curve models. Both cPTEN scoring by the pathologist and quantification of PTEN loss by AI (high-risk AI-qPTEN) were significantly associated with shorter metastasis-free survival (MFS) in univariable analysis (cPTEN hazard ratio [HR], 1.54; CI, 1.07-2.21; P = .019; AI-qPTEN HR, 2.55; CI, 1.83-3.56; P < .001). In multivariable analyses, AI-qPTEN showed a statistically significant association with shorter MFS (HR, 2.17; CI, 1.49-3.17; P < .001) and recurrence-free survival (HR, 1.36; CI, 1.06-1.75; P = .016) when adjusting for relevant postsurgical clinical nomogram (Cancer of the Prostate Risk Assessment [CAPRA] postsurgical score [CAPRA-S]), whereas cPTEN does not show a statistically significant association (HR, 1.33; CI, 0.89-2; P = .2 and HR, 1.26; CI, 0.99-1.62; P = .063, respectively) when adjusting for CAPRA-S risk stratification. More importantly, AI-qPTEN was associated with shorter MFS in patients with favorable pathological stage and negative surgical margins (HR, 2.72; CI, 1.46-5.06; P = .002). Workflow also demonstrated enhanced clinical utility in decision curve analysis, more accurately identifying men who might benefit from adjuvant therapy postsurgery. This study demonstrates the clinical value of an affordable and fully automated AI-powered PTEN assessment for evaluating the risk of developing metastasis or disease recurrence after radical prostatectomy. Adding the AI-qPTEN assessment workflow to clinical variables may affect postoperative surveillance or management options, particularly in low-risk patients.

Distinguishing between true indolent and potentially life-threatening prostate cancer is challenging in tumours displaying clinicopathologic features associated with low or intermediate risk of relapse. Several somatic DNA copy number alterations (CNAs) have been identified as potential prognostic biomarkers, but the standard cytogenetic method to assess them has a limited multiplexing capability. Multiplex ligation-dependent probe amplification (MLPA) targeting 14 genes was optimised to survey 448 tumours of patients with low or intermediate risk (Grade Group 1–3, Gleason score ≤7) who underwent radical prostatectomy. A 6-gene CNA classifier was developed using random survival forest and Cox proportional hazard modelling to predict biochemical recurrence. The classifier score was significantly associated with biochemical recurrence after adjusting for standard clinicopathologic variables and the known prognostic index CAPRA-S score with a hazard ratio of 2.17 and 1.80, respectively (n = 406, P < 0.01). The prognostic value of this classifier was externally validated in published CNA data from three radical prostatectomy cohorts and one radiation therapy pre-treatment biopsy cohort. The 6-gene CNA classifier generated by a single MLPA assay compatible with the small quantities of DNA extracted from formalin-fixed paraffin-embedded (FFPE) tissue specimens has the potential to improve the clinical management of patients with low or intermediate risk disease.

Complex high-dimensional datasets that are challenging to analyze are frequently produced through '-omics' profiling. Typically, these datasets contain more genomic features than samples, limiting the use of multivariable statistical and machine learning-based approaches to analysis. Therefore, effective alternative approaches are urgently needed to identify features-of-interest in '-omics' data. In this study, we present the molecular feature selection tool, a novel, ensemble-based, feature selection application for identifying candidate biomarkers in '-omics' data. As proof-of-principle, we applied the molecular feature selection tool to identify a small set of immune-related genes as potential biomarkers of three prostate adenocarcinoma subtypes. Furthermore, we tested the selected genes in a model to classify the three subtypes and compared the results to models built using all genes and all differentially expressed genes. Genes identified with the molecular feature selection tool performed better than the other models in this study in all comparison metrics: accuracy, precision, recall, and F1-score using a significantly smaller set of genes. In addition, we developed a simple graphical user interface for the molecular feature selection tool, which is available for free download. This user-friendly interface is a valuable tool for the identification of potential biomarkers in gene expression datasets and is an asset for biomarker discovery studies.

Background: Pelvic nodal metastasis in prostate cancer impacts patient outcome negatively. Objective: To explore tumor-infiltrating immune cells as a potential predictive tool for regional lymph node (LN) metastasis. Design, setting, and participants: We applied multiplex immunofluorescence and targeted transcriptomic analysis on 94 radical prostatectomy specimens in patients with (LN+) or without (LN) pelvic nodal metastases. Both intraepithelial and stromal infiltrations of immune cells and differentially expressed genes (mRNA and protein levels) were correlated with the nodal status. Outcome measurements and statistical analysis: The identified CD4 effector cell signature of nodal metastasis was validated in a comparable independent patient cohort of 184 informative cases. Patient outcome analysis and decision curve analysis were performed with the CD4 effector cell density-based signature. Results and limitations: In the discovery cohort, both tumor epithelium and stroma from patients with nodal metastasis had significantly lower infiltration of multiple immune cell types, with stromal CD4 effector cells highlighted as the top candidate marker. Targeted gene expression analysis and confirmatory protein analysis revealed key alteration of extracellular matrix components in tumors with nodal metastasis. Of note, stromal CD4 immune cell density was a significant independent predictor of LN metastasis (odds ratio [OR] = 0.15, p = 0.004), and was further validated as a significant predictor of nodal metastasis in the validation cohort (OR = 0.26, p < 0.001). Conclusions: Decreased T-cell infiltrates in the primary tumor (particularly CD4 effector cells) are associated with a higher risk of LN metastasis. Future evaluation of CD4-based assays on prostate cancer diagnostic biopsy materials may improve selection of at-risk patients for the treatment of LN metastasis. Patient summary: In this report, we found that cancer showing evidence of cancer metastasis to the lymph nodes tends to have less immune cells present within the tumor. We conclude that the extent of immune cells present within a prostate tumor can help doctors determine the most appropriate treatment plan for individual patients. (C) 2021 Published by Elsevier B.V.

Abstract BACKGROUND: In recent years AI and deep learning have transformed the ability to use large amounts of medical data to augment diagnosis and prognosis processes for cancer. For developing AI methodology, histopathologic assessment serves as the gold standard “labels”, enabling investigators to finely map (or annotate) biologically and clinically important features. Yet correlating high dimensional data (radiomic, morphometric, genomic, metabolomic, etc.) with expert histopathologic diagnosis for dataset generation remains a major challenge. CHALLENGE: Traditionally, labels have been extracted from a snapshot that contains all of the annotation layers overlaid on the original tissue through image processing techniques. This implies the use of distinct colors for annotation, which severely constrain the number of possible labels. Particularly, this is most noticeable for heterogeneous tissues like prostate that require complex annotation. Furthermore, the resolution with which the labels can be mapped is limited by the area of the extracted region. OBJECTIVE: Here we present a workflow for pathology-oriented dataset generation for AI studies that is compatible with standard annotation platforms, and addresses these limitations. We introduce a detailed multi-grade and multi-scale annotation protocol for prostate biopsies. The proposed method is capable of exporting labels as independent layers (representing specific grades of the pathology), and resampling them to the desired resolution. METHODS: A collection of 38 prostate biopsy sections from 19 patients fixed on slides were used. The proposed grading annotation protocol is based on the spatial distribution of cancer cells. Nine layers of annotation were considered depicting stroma, benign tissue, low grade (Gleason pattern 3) and high grade cancer (Gleason patterns 4 and 5), two mixed cancer patterns, prostatic intraepithelial neoplasia (PIN), intraductal carcinoma (IDC), and artifact. The coordinates of the annotation boundaries are post-processed and combined into a label image containing all 9 pathological classes. The metabolomic profiles of the prostate biopsies acquired by desorption electrospray ionization (DESI) is considered for data features in this study. The generated image labels are therefore spatially registered to corresponding DESI data of each slide. RESULTS: The generated dataset through proposed method is used in the application of prostate cancer detection. The dataset is validated through qualitative visualization and quantitative analysis. High correlation is observed between label images of the slides and unsupervised linear representation of corresponding DESI spectra. The pixel-based supervised identification of tissue types based on the DESI also shows high accuracy. CONCLUSION: The proposed digitized pathology annotation protocol and dataset generation workflow is compatible with AI oriented cancer research and is capable of handling large number of pathological classes and high dimensional imaging modalities. Citation Format: Amoon Jamzad, Tamara Jamaspishvili, Rachael Iseman, Martin Kaufmann, David Berman, Parvin Mousavi. An efficient digitized annotation platform for pathology-oriented dataset generation in AI research [abstract]. In: Proceedings of the AACR Virtual Special Conference on Artificial Intelligence, Diagnosis, and Imaging; 2021 Jan 13-14. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(5_Suppl):Abstract nr PO-005.

209 Background: Histopathologic investigation of diagnostic prostate biopsies both confirms the presence of disease and estimates its potential for distal spread via tumour grade. The accuracy of biopsy grading is limited by intra-tumoral heterogeneity, inter-observer variability, and other factors. To improve risk stratification at the time of diagnosis, we sought to create objective molecular biomarkers of radical prostatectomy grade that are resistant to sampling error and should be useful when applied to biopsy tissue. Methods: We developed and validated a robust objective biomarker of prostate cancer grade using pathologic grading of prostatectomy tissues as the gold standard. We created training (333 patients) and validation (202 patients) cohorts of Cancer of the Prostate Risk Assessment (CAPRA) low- and intermediate-risk prostate cancer patients. To address intra-tumoral heterogeneity, each tumor was sampled at two locations. We profiled the abundance of 342 mRNAs complemented by 100 canonical DNA copy number aberration loci (CNAs) and 14 hypermethylation events. Using the training cohort with cross-validation, we evaluated models for training classifiers of pathologic Grade Group ≥2, Restricting to strategies resulting in true negative rates ≥0.5, true positive (TP) rates ≥0.8, we selected two strategies to train classifiers, PRONTO-e and PRONTO-m. Results: The PRONTO-e classifier comprises 353 mRNA and CNA features, while the PRONTO-m classifier comprises 94 mRNA, CNA, methylation and clinical features. Both classifiers (PRONTO-e, PRONTO-m) validated in the independent cohort, with respective TP rates of 0.809 and 0.760, false positive rates of 0.429 and 0.262, F1 scores of 0.709 and 0.724, and AUCs of 0.792 and 0.818. Conclusions: Two classifiers were developed and validated in separate cohorts, each achieved excellent performance by integrating different types of molecular data. Implementation of classifiers with these performance characteristics could markedly improve current active surveillance approaches without increasing patient morbidity and may help better inform patients on their individual need for definitive therapy versus active surveillance.