hupo-2024-eposter-2-6667f1b761d39446700467.jpg

hupo-2024-eposter-logo-6667fbf57196e975576823.svg

hupo-2024-eposter-6667f277b8bac153483681.jpg

23rd Human Proteome Organization World Congress

REAL: The principle and approach for clinical multicenter proteomic tumor analysis

Professor

Keynote speaker

Keynote lecture

Prof.

(Phospho)proteomics for personalized precision oncology

Keynote Lecture

Cancer is a heterogeneous disease characterized by aberrant cellular signaling. Hyperactive kinase activity is one of the drivers of tumorigenesis and cancer progression and results in altered phosphorylation abundance and activity of downstream substrates. A prime prerequisite for personalized treatment is that kinase activity inference can be based on a single (patient) sample. This is pivotal in a clinical setting with the aim to select the best kinase inihibitor treatment for individual patients.To this end, phosphoproteomics coupled to INKA analysis has emerged as an advantageous approach that achieves an optimized ranking of inferred kinase activities in single samples (www.inkascore.org) (1-5). INKA integrates kinase phosphorylation and substrate phosphorylation to infer kinase activity. The output is a ranked kinase activity bar graph and an individualized kinase-substrate network that can be used by biologists and clinicians to extract actionable information from modern deep comprehensive (patient) phosphoproteomic data.In my lecture, I will present our progress in (phospho)proteomics profiling and INKA of xenograft tumor models and tumor needle biopsies collected as part of side-studies in precision medicine clinical trials (IMPACT, colorectal cancer patients treated with anti-EGFR therapy; REPOSIT, melanoma patients treated with anti-BRAF-MEK combination therapy; DRUP-Lenvatinib cohort, pan-cancer). Our findings show that needle biopsy protein yields are compatible with phosphoproteomics (~200 mg protein) and a large fraction of the samples is also amendable to pTyr-phosphoproteomics that we recently down-scaled to 500 mg protein input. The analyses offer explanation for (lack of) drug response and uncover actionable resistance mechanisms.Ultimately, prospective clinical trials that include drug selection based on patient phosphoproteomic data coupled to individualized kinase and pathway activity inference, will show the added value of phosphoproteomics (combined with genomics) for precision medicine.Beekhof R, Henneman AA, Jimenez CR et al.,. INKA, an integrative data analysis pipeline for phosphoproteomic inference of active kinases. Mol Syst Biol. 2019 May 24;15(5):e8981.Cordo' V, et al., Phosphoproteomic profiling of T cell acute lymphoblastic leukemia reveals targetable kinases and combination treatment strategies. Nat Commun. 2022 Feb 25;13(1):1048. Vallés-Martí A, et al. Phosphoproteomics guides effective low-dose drug combinations against pancreatic ductal adenocarcinoma. Cell Rep. 2023 Jun 1;42(6):112581.Beekhof R, et al., Phosphoproteomics of patient-derived xenografts identifies targets and markers associated with sensitivity and resistance to EGFR blockade in colorectal cancer. Sci Transl Med. 2023 Aug 16;15(709):eabm3687.Vallés-Martí A, et al. Kinase activities in pancreatic ductal adenocarcinoma with prognostic and therapeutic avenues. Mol Oncol. 2024 Aug;18(8):2020-2041.

Federated learning enables cancer subtyping by proteomics

Clinical Proteomics

Introduction:The availability of large quantities of online data is underpinning the rapid development of many artificial intelligence (AI) applications. However, in biological and healthcare settings AI applications face challenges, including data privacy and security. Federated learning (FL), a distributed machine learning method, enables analysis of large biomedical datasets while ensuring data privacy, confidentiality and regulatory compliance. In this approach, local models are trained using data held inside several participating sites, with the data remaining private behind local firewalls. Local model updates are then aggregated in the server to improve the global model.Methods: We obtained 30 distinct tissue sample cohorts, including a pan-cancer cohort (N = 1,260) from biobanks in Australia, with the remainder (N = 6,265) being largely single-cancer cohorts from centres in North America, Europe and Australia. We quantified 9,051 proteins across 7,525 tissue samples encompassing 57 cancer subtypes, using data-independent acquisition mass spectrometry (DIA-MS). From this we developed ProCanFL, a deep learning-based FL model that is trained on multiple simulated local sites using samples from the pan-cancer cohort and 90% of the 29 other cohorts. We simulated four local sites. The first site was fixed and housed the proteomic data and histopathology labels of the pan-cancer cohort, while sites 2-4 contained data from random combinations of the 29 cohorts. This randomization process was repeated 10 times to create 10 cohort allocations. A local model was trained at each site and the global model incorporated parameter updates from the local models.Results: We used a hold-out test set (10% of the 29 cohorts) to evaluate the performance of individual local models in predicting 15 cancer subtypes for which a minimum number of samples were represented in the proteomic dataset. The average accuracy across the four sites, evaluated on the unseen samples in the test set was 0.68±0.14. After application of ProCanFL, the global model had a significantly enhanced predictive performance, achieving an accuracy of 0.97±0.01.Conclusion: We demonstrated that FL enables a global model to be trained using local model updates from multiple decentralized sites containing private data, achieving high performance in cancer subtype prediction. Our study underscores the potential role of FL in the biomedical domain, where data sharing among global consortium members is often limited by several barriers, including country-specific data regulations. Further, the prediction accuracy from the FL model can be improved as more local sites and data are added. By harnessing the full wealth of large global datasets while keeping local data behind firewalls, FL offers a promising solution for developing omic foundation models and advancing the future of digital health.

Oral presentation

Proteogenomic analysis revealed prognosis biomarker and therapeutic targets in sarcomas

Backgrounds:Sarcomas are rare mesenchymal malignant tumors. More than 100 histological subtypes have been characterized, and most of them have limited treatment options. Genomic analyses for sarcomas were reported, but proteomic analyses leading to diagnosis and treatment have not yet been conducted.Objective:The purpose of this study was to develop diagnostic or prognostic biomarkers and discover therapeutic targets in sarcomas.Methods:Proteogenomic analyses were performed on frozen tissues and cell lines derived from approximately 1,000 patients with 12 histologically different sarcomas. All materials including clinical and pathological information were obtained from the National Cancer Center Biobank, Japan. Proteins were extracted, purified with single-pot solid-phase-enhanced sample preparation (SP3), digested by Trypsin/Lys-C Mix, and subjected to the Orbitrap Astral mass spectrometer for quantitative proteomics using Data-Independent Acquisition (DIA). We integrated proteomic data and previously obtained genomic and transcriptomic data.Results:Sarcomas were classified into various proteomic clusters based on the expression pattern of 8000-9000 proteins. Pathway enrichment analysis revealed the specific molecular pathways unique to individual clusters. The integrated analysis showed correlations between genomic and proteomic data. Comparative analysis of proteomic and transcriptomic profiles highlighted the proteomic-specific features for optimal risk stratification. We identified proteins common to the poor prognosis group as prognostic biomarkers, and molecular pathways that could be therapeutic targets for molecular target drugs.Conclusion:This proteogenomic study revealed the utility of proteogenomics for identifying molecular subgroups with implications for risk stratification and therapy selection. We confirmed that this study provided a rich resource for future sarcoma research. We will conduct further analyses for the development of precision medicine for sarcomas.

Integrating AI-based Digital Pathology and Proteome Profiling to Uncover Melanoma Progression and Heterogeneity

Melanoma, the deadliest form of skin cancer, is one of the fastest-growing cancers in the world. The marked molecular and morphological heterogeneity of this disease is directly connected to its aggressive behavior, progression, and treatment resistance. In this study, we integrate AI-based digital pathology and proteome profiling to deeply characterize histopathological features and melanoma subclones, and determine their molecular fingerprint in different melanoma cases to uncover the mechanism linked with progression.Using artificial intelligence approaches and high-resolution HE images, tumor and stroma areas, histopathological components, and melanoma subclones were identified, further isolated by laser microdissection, and analyzed by a proteomics approach. Over 100 FFPE samples were investigated, corresponding to primary tumors (n=84) and paired metastases (n=20). The samples were selected from patients who experienced disease recurrence within 5 years, with metastases at the moment of diagnosis, and those without any progression even after 5 years of the initial diagnosis. By applying this approach, we investigated the whole progression of melanoma from early-stage to advanced metastatic cases, decoding the pathways involved in the disease progression.The AI-based digital pathology results unveiled histopathological signatures in tumor cells and their surroundings, encompassing melanoma subclones and stromal components. Determining the proteomic signature of these areas revealed dysregulation of mitochondrial function and immune response-related pathways as the most significant drivers of progression. Proliferative pathways and histidine metabolism were up- and down-regulated, respectively, in line with the advancement of the disease. Additionally, suppression of immune system response in the tumor microenvironment of recurrent and metastatic melanomas underscores the importance of immunotherapies to improve patient outcomes in these groups of patients. The analyses of different melanoma subclones revealed molecular heterogeneity linked to metastatic melanoma and evidenced the critical role of mitochondria and histidine metabolism in melanoma progression, opening an opportunity to identify new druggable targets for therapy. Additionally, our study identified morphological markers associated with a high risk of melanoma recurrence and progression. Specifically, the presence of microerosion, budding formations, and melanocytic atypia, along with the absence of regression areas, were found to be significant risk factors for melanoma progression.Altogether, our findings highlighted that integrating AI-based digital pathology and proteome profiling is a powerful strategy to map melanoma progression, identify new druggable targets for treatment, and determine markers of recurrence and progression, which may illuminate new early-stage melanoma prognostic factors and guide a new era of personalized therapy.

Plenary hall

Session

Proteomics in oncology

Session
Proteomics in oncology

Termin

Vorsitzende

Programm

SessionProteomics in oncology

Termin

Vorsitzende

Programm

Session
Proteomics in oncology