Biology subjects

Dezso, Z.

Publications and source records attributed to Dezso, Z..

3 recordsLinked to original sources

bioRxiv2025

INSIGHT: In Silico Drug Screening Platform using Interpretable Deep Learning Network

The large-scale multiplexed drug screening platforms like PRISM and GDSC facilitate the screening of drug treatments over 1,000 cancer cell lines. The cancer cell lines are well characterized by multiomics screening in CCLE and DepMap, enabling the application of AI and machine learning techniques to study the association between drug sensitivity and the underlying molecular profiles. The large scale and variety of data modalities enabled us to build an interpretable deep learning framework, INSIGHT, integrating the multiomics data and the drugs molecular structure to predict drug response. We trained our model on the PRISM screen for single treatments and on the DrugComb screen database for combination treatments. Our method enables the in-silico extension of current screens by predicting drug response in cancer cell lines not included in the screen, as well as the drug response to novel single agent or combination therapy by leveraging the drugs molecular structure. Furthermore, the deep learning framework was built to enable biological interpretation. The connections between the hidden layers of the neural network incorporate prior biological knowledge such as signaling pathways. This enables the model, in addition to predicting the drug sensitivity profiles, to prioritize the pathways predictive of drug response and identify pathways related to the mechanism of action (MOA) and potential off target effects of novel drugs. The evaluation of our model using cross-validation on the PRISM and DrugComb dataset showed an improved performance compared to previously developed biologically informed deep learning methods and traditional state of the art machine learning methods like elastic-net and XGBoost. We illustrated with examples the value of incorporating biological knowledge into INSIGHT by relating the pathway activity of the predictive models to the MOA.

bioinformatics↗

bioRxiv2022

Machine Learning Reveals Genetic Modifiers of the Immune Microenvironment of Cancer

Heritability in the immune tumor microenvironment (iTME) has been widely observed, yet remains largely uncharacterized and systematic approaches to discover germline genetic modifiers of the iTME still being established. Here, we developed the first machine learning approach to map iTME modifiers within loci from genome-wide association studies (GWAS) for breast cancer (BrCa) incidence and outcome. A random forest model was trained on a positive set of immune-oncology (I-O) targets using BrCa and immune phenotypes from genetic perturbation studies, comparative genomics, Mendelian genetics, and colocalization with autoimmunity and inflammatory disease risk loci. Compared with random negative sets, an I-O target probability score was assigned to the 1,362 candidate genes in linkage disequilibrium with 155 BrCa GWAS loci. Pathway analysis of the most probable I-O targets revealed significant enrichment in drivers of BrCa and immune biology, including the LSP1 locus associated with BrCa incidence and outcome. Quantitative cell type-specific immunofluorescent imaging of 1,109 BrCa patient biopsies revealed that LSP1 expression is restricted to tumor infiltrating leukocytes and correlated with BrCa patient outcome (HR = 1.73, p < 0.001). The human BrCa patient-based genomic and proteomic evidence, combined with phenotypic evidence that LSP1 is a negative regulator of leukocyte trafficking, prioritized LSP1 as a novel I-O target. Finally, a novel comparative mapping strategy using mouse genetic linkage revealed TLR1 as a plausible therapeutic candidate with strong genomic and phenotypic evidence. Collectively, these data demonstrate a robust and flexible analytical framework for functionally fine-mapping GWAS risk loci to identify the most translatable therapeutic targets for the associated disease.

cancer biology↗

bioRxiv2022

TCGADEPMAP -- Mapping Translational Dependencies and Synthetic Lethalities within The Cancer Genome Atlas

The Cancer Genome Atlas (TCGA) has yielded unprecedented genetic and molecular characterization of the cancer genome, yet the functional consequences and patient-relevance of many putative cancer drivers remain undefined. TCGADEPMAP is the first hybrid map of translational tumor dependencies that was built from machine learning of gene essentiality in the Cancer Dependency Map (DEPMAP) and then translated to TCGA patients. TCGADEPMAP captured well-known and novel cancer lineage dependencies, oncogenes, and synthetic lethalities, demonstrating the robustness of TCGADEPMAP as a translational dependency map. Exploratory analyses of TCGADEPMAP also unveiled novel synthetic lethalities, including the dependency of PAPSS1 driven by loss of PAPSS2 which is collaterally deleted with the tumor suppressor gene PTEN. Synthetic lethality of PAPSS1/2 was validated in vitro and in vivo, including the underlying mechanism of synthetic lethality caused by the loss of protein sulfonation that requires PAPSS1 or PAPSS2. Moreover, TCGADEPMAP demonstrated that patients with predicted PAPSS1/2 synthetic lethality have worse overall survival, suggesting that these patients are in greater need of drug discovery efforts to target PAPSS1. Other map "extensions" were built to capture unique aspects of patient-relevant tumor dependencies using the flexible analytical framework of TCGADEPMAP, including translating gene essentiality to drug response in patient-derived xenograft (PDX) models (i.e., PDXEDEPMAP) and predicting gene tolerability within normal tissues (GTEXDEPMAP). Collectively, this study demonstrates how translational dependency maps can be used to leverage the rapidly expanding catalog of patient genomic datasets to identify and prioritize novel therapeutic targets with the best therapeutic indices.

cancer biology↗