Integrated spatial morpho-transcriptomics predicts functional traits in pancreatic cancer

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2025, Номер unknown

Опубликована: Март 14, 2025

Analyses of patient-derived cell lines have greatly enhanced discovery molecular biomarkers and therapeutic targets. However, characterization cellular morphological properties is limited. We studied morphologies human pancreatic adenocarcinoma (PDAC) their associations with drug sensitivity, gene expression, functional properties. By integrating live spatial mRNA imaging, we identified KRAS inhibitor-induced changes specific for drug-resistant cells that correlated expression changes. then categorized a large panel PDAC into (e.g., polygonal, irregular, spheroid) organizational tightly aggregated, multilayered, dispersed) subtypes found differences in targeting potential, metastatic proclivity. In tissues, prognostic signatures associated distinct cancer organization patterns. summary, highlight the potential information rapid, cost-effective assays to aid precision oncology efforts leveraging vitro models tissues.

Язык: Английский

Unifying Genetic and Chemical Perturbagen Representation through a Hybrid Deep Learning Framework

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2025, Номер unknown

Опубликована: Фев. 7, 2025

The integration of genetic and chemical perturbations has driven transformative advances in elucidating cellular mechanisms accelerating drug discovery. However, the lack a unified representation for diverse perturbagen types limits comprehensive analysis joint modeling multi-domain perturbation agents (molecular cause space) their resulting phenotypes (phenotypic effect spaces). Here, we present UniPert, hybrid deep learning framework that encodes perturbagens into shared semantic space. UniPert employs tailored encoders to address inherent molecular-scale differences across leverages contrastive with experiment-driven compound-target interactions bridge these domains. Extensive experiments validate UniPert’s versatility application. generated representations effectively capture hierarchical pharmacological relationships perturbagens, facilitating annotations understudied targets compounds. can be plugged advanced frameworks enhance performance both outcome prediction tasks. Notably, paves way cross-domain modeling, driving novel genetic-to-chemical transfer paradigm, boosting context-specific silico screening efficiency development personalized therapies.

Язык: Английский

A Framework for Autonomous AI-Driven Drug Discovery

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2024, Номер unknown

Опубликована: Дек. 20, 2024

Abstract The exponential increase in biomedical data offers unprecedented opportunities for drug discovery, yet overwhelms traditional analysis methods, limiting the pace of new development. Here we introduce a framework autonomous artificial intelligence (AI)-driven discovery that integrates knowledge graphs with large language models (LLMs). It is capable planning and carrying out automated programs while providing details its research strategy, progress, supporting points, enabling thorough assessment methods findings. At heart this lies “focal graph” - novel construct harnesses centrality algorithms to distill vast, noisy datasets into concise, transparent, data-driven hypotheses. By high-throughput search result interpretation, such could be used execute massive numbers searches, identify patterns across complex, diverse datasets, prioritize actionable hypotheses at scale speed unachievable by human researchers alone. We demonstrate even small- applications approach can yield novel, transparent insights relevant multiple stages process present prototype system autonomously executing multi-step target workflow. focal graph described here, automation it enables, represents promising path forward: towards deeper understanding mechanisms underlying disease true acceleration development therapeutics. Graphical

Язык: Английский