Periodic Reporting for period 3 - iPC (individualizedPaediatricCure: Cloud-based virtual-patient models for precision paediatric oncology)
Periodo di rendicontazione: 2022-01-01 al 2023-11-30
The treatment of paediatric cancers poses multifaceted challenges. This situation leads to a lack of targeted therapies, inaccurate risk stratification, over-treatment, and compromised quality of life for young patients. Additionally, the focus on chemotherapy brings forth long-term side effects. Addressing these challenges is critical as cancer remains the leading cause of death by disease among children. Improved risk prediction and tailored therapies are pivotal to optimize treatment outcomes while reducing adverse effects. By bridging fragmented knowledge and leveraging advanced technologies, the project pioneers the field of personalized medicine. This effort aligns with the societal need for precise, effective treatments, ensuring better survival rates and enhanced quality of life for young cancer patients. The iPC consortium has embarked on a initiative to revolutionize paediatric cancer care. The overarching goal has been the integration of diverse datasets and cutting-edge models, both knowledge-based and AI-driven, to develop a virtual patient framework. This framework will empower clinicians with computational tools for personalized diagnostics and treatment recommendations, through: Data Integration, Model Development, Virtual Patient Framework. The project not only seeks to fill critical gaps in paediatric cancer care but also aims to leverage existing knowledge and technological advancements to benefit thousands of children diagnosed with cancer.
WP1: iPC has generated a great deal of data for analysis. These include scRNA-Seq data of tumours, testing data for deconvolution efforts, patient data to train new analysis models, incl. to distinguish cancer from non-cancer cells, and proteogenomic data to test models that predict drug response, as well as large-scale perturbation data for testing predictive models. In addition, iPC has developed methods to harmonise RNA-Seq data and applied these methods on a large collection of paediatric cancer datasets.
WP2: The iPC Data Access Framework prototype has been deployed to the BSC cloud and integrated with the rest of the platform components. The iPC Data Access Framework components have been adapted in order to interoperate with external systems leveraging the GA4GH Passports/Visas specification. The iPC Computational Platform has successfully been integrated with the EGA Data Access Framework.
WP3: Progress was made in structure-based deep learning to predict protein stability effects. WP3 also focused on understanding CAR-T immunotherapies in liver cancer patients and uncovered molecular differences and depletion patterns in CAR-T cells between adult and paediatric patients. BIODICA was also refined for matrix factorisation analysis and ROMA was improved for the assessment of gene set activity, especially in multi-omics datasets. Also noteworthy was the development of the multiAffinity tool for the molecular characterisation of hepatoblastomas and the development of the iPC Data Access Framework.
WP4: WP4 prepared for the high-quality reference PPI networks for network-based analysis of ICA metagenes in the last section to also link the networks to the results. The WP4 partners also developed a method for defining network modules from multi-omics datasets and applied it to a multi-omics neuroblastoma dataset.
WP5: WP5 implemented and tested a neural network as an edge classifier. It was shown that the neural network for graph selection can "decode" samples from graphs in an information-theoretically optimal way. The team also worked on the introduction of a new network architecture for transforming data matrices in arbitrary form and also on the construction and validation of methods for predicting the effects of radiotherapy on paediatric cancers with identification of regulatory elements and interactions.
WP6: WP6 focused on optimising the model parameters and improving the accuracy of the model. The model was extended to include additional (off-)targets of ceritinib and the automatic reporting process was refined. The effect of the chemotherapeutic cancer drug on inducing an inflammatory state leading to the generation of a humoral and a cytotoxic immune response against specific tumour-associated antigens was implemented. This resulted in an improvement of the simulation with respect to the different cell populations modelled for the specific case of hepatoblastoma.
WP7: WP7 developed and validated three new methods for computational deconvolution. One aimed at performing inference from scRNA-seq data to map heterogeneity, one to perform computational deconvolution and one to predict miRNA targets using multi-omics data. The performance of the methods were compared to other computational deconvolution methods and demonstrated to outperform these across different datasets.
WP8: A computational method for predicting small molecule inhibitors of cancer cell growth was developed and molecular targets for high-risk AML and HB patients were identified. The team combined gene expression and genomic variants to create sample-specific networks and find causal links between transcription factor activities and upstream nodes. The partners have developed a metabolic model of intratumoural heterogeneity in Ewing sarcoma tumours and specialised both a liver and HB metabolic model for performing differential flux balance analysis.
WP9: WP9 worked on many different lines of research, including the first study to validate an auto-encoder model for predicting both developmental progression and disease status of single cells. The improved CAR-T design has shown improved efficacy in patients. The effects of CDK9 inhibitors in a xenograft mouse model were also analysed and the partners have also validated that the developmental course of bone marrow cells of the myeloid lineage can be recorded and that it can be used to determine at which point in the developmental course the leukaemia occurs.
WP2: The iPC Data Access Framework prototype has been deployed to the BSC cloud and integrated with the rest of the platform components. The iPC Data Access Framework components have been adapted in order to interoperate with external systems leveraging the GA4GH Passports/Visas specification. The iPC Computational Platform has successfully been integrated with the EGA Data Access Framework.
WP3: Progress was made in structure-based deep learning to predict protein stability effects. WP3 also focused on understanding CAR-T immunotherapies in liver cancer patients and uncovered molecular differences and depletion patterns in CAR-T cells between adult and paediatric patients. BIODICA was also refined for matrix factorisation analysis and ROMA was improved for the assessment of gene set activity, especially in multi-omics datasets. Also noteworthy was the development of the multiAffinity tool for the molecular characterisation of hepatoblastomas and the development of the iPC Data Access Framework.
WP4: WP4 prepared for the high-quality reference PPI networks for network-based analysis of ICA metagenes in the last section to also link the networks to the results. The WP4 partners also developed a method for defining network modules from multi-omics datasets and applied it to a multi-omics neuroblastoma dataset.
WP5: WP5 implemented and tested a neural network as an edge classifier. It was shown that the neural network for graph selection can "decode" samples from graphs in an information-theoretically optimal way. The team also worked on the introduction of a new network architecture for transforming data matrices in arbitrary form and also on the construction and validation of methods for predicting the effects of radiotherapy on paediatric cancers with identification of regulatory elements and interactions.
WP6: WP6 focused on optimising the model parameters and improving the accuracy of the model. The model was extended to include additional (off-)targets of ceritinib and the automatic reporting process was refined. The effect of the chemotherapeutic cancer drug on inducing an inflammatory state leading to the generation of a humoral and a cytotoxic immune response against specific tumour-associated antigens was implemented. This resulted in an improvement of the simulation with respect to the different cell populations modelled for the specific case of hepatoblastoma.
WP7: WP7 developed and validated three new methods for computational deconvolution. One aimed at performing inference from scRNA-seq data to map heterogeneity, one to perform computational deconvolution and one to predict miRNA targets using multi-omics data. The performance of the methods were compared to other computational deconvolution methods and demonstrated to outperform these across different datasets.
WP8: A computational method for predicting small molecule inhibitors of cancer cell growth was developed and molecular targets for high-risk AML and HB patients were identified. The team combined gene expression and genomic variants to create sample-specific networks and find causal links between transcription factor activities and upstream nodes. The partners have developed a metabolic model of intratumoural heterogeneity in Ewing sarcoma tumours and specialised both a liver and HB metabolic model for performing differential flux balance analysis.
WP9: WP9 worked on many different lines of research, including the first study to validate an auto-encoder model for predicting both developmental progression and disease status of single cells. The improved CAR-T design has shown improved efficacy in patients. The effects of CDK9 inhibitors in a xenograft mouse model were also analysed and the partners have also validated that the developmental course of bone marrow cells of the myeloid lineage can be recorded and that it can be used to determine at which point in the developmental course the leukaemia occurs.
The iPC project has significantly advanced the field by overcoming the complexities in integrating diverse patient data modalities for personalized disease models. iPC's collaborative expertise spanning computational modeling and clinical patient management resulted in the creation of a wide range of software tools usable by clinicians. These tools enable clinicians and healthcare professionals to efficiently translate intricate multi-level patient data into actionable insights. The integration of the multiple data modalities collected for each patient holds the key to precise personalised models. iPC successfully developed computational models and workflows for systematic integration of vast quantitative data, utilizing diverse modeling approaches and high-performance computing. Furthermore, the project integrated advanced mechanistic, statistical, and AI models into a virtual patient framework, facilitating deep molecular analyses and treatment recommendations. Novel methodologies for network reconstruction, personalized immune system modeling, automated Comprehensive Molecular Tumour Analyses reports, and a secure online platform were achieved milestones. The project's innovation and technological advancement has pushed the state of the art, bridging the gap between complex data and practical clinical applications, to make substantial socio-economic impacts by enhancing personalised healthcare and improving outcomes for paediatric cancer patients.