[R35] Decoding Complexity: Representation Learning for Disease Mechanisms Across Biological Contexts

Ente: National Institute of General Medical Sciences

Scadenza: 2031-04-30

Importo max: 420.600 EUR

Paese: US

Descrizione

Abstract Diseases arise from dynamic molecular and cellular changes that span tissues, cell types, and patients. Yet our ability to study these processes remains constrained by incomplete data: many tissues and disease states cannot be sampled, and single-cell datasets, though powerful, are sparse and biased. As a result, most studies provide only fragmented views of disease, obscuring the mechanisms that drive progression, heterogeneity, and response to therapy. This proposal develops new computational frameworks that address these gaps by learning interpretable data representations that integrate bulk and single-cell transcriptomic data across tissues, diseases, and technologies. These representations will enable us to infer missing data, resolve cell-type-specific changes, and reconstruct trajectories of disease progression. By leveraging large heterogeneous datasets, our models will provide dynamic, mechanistic insight into how diseases initiate, evolve, and diverge across individuals. Direction 1 focuses on resolving cell-to-trait associations. Current approaches depend on hand-picked single-cell reference datasets, which is a tedious process to perform at scale across multiple tissues, diseases, and populations. We will instead build multimodal representations that integrate expression data with metadata, enabling automatic and interpretable inference of cell type proportions and expression across thousands of bulk samples. This will generate a foundational resource for studying how cell composition and state changes underlie the development of complex diseases. Direction 2 addresses the challenge of inferring missing expression data. Diseases often emerge across multiple tissues and progress over long periods, but our observations are incomplete because tissue samples cannot always be obtained due to ethical considerations. Additionally, many diseases remain asymptomatic until later stages, making early sampling particularly challenging. To overcome this, we will develop methods that leverage similarities across tissues and diseases to impute missing tissue-disease pairs and reconstruct disease trajectories from healthy to diseased states. All of these methods will use interpretable models applied to large, heterogeneous datasets, enabling researchers to easily study disease mechanisms across large cohorts that span multiple tissues and populations. This work will uncover insights that are not directly observable, generate broadly reusable community resources, and establish a robust computational framework for investigating disease biology. Istituzione: UNIVERSITY OF COLORADO DENVER PI: Natalie Rose Davidson Progetto: 1R35GM165442-01

Settori: National Institute of General Medical Sciences

Vai al bando originale

Registrati gratis su Bandolo per trovare bandi compatibili con la tua azienda.