Biology subjects

Knijnenburg, T.

Publications and source records attributed to Knijnenburg, T..

3 recordsLinked to original sources

bioRxiv2017

Mining the forest: uncovering biological mechanisms by interpreting Random Forests

Biological datasets are large and complex. Machine learning models are therefore essential to capture relationships in the data. Unfortunately, the inferred complex models are often difficult to understand and interpretation is limited to a list of features ranked on their importance in the model.\n\nWe propose a computational approach, called Foresight, that enables interpretation of the patterns uncovered by Random Forest models trained on biological datasets. Foresight exploits the correlation structure in the data to uncover relevant groups of features and the interactions between them. This facilitates interpretation of the computational model and can provide more detailed insight in the underlying biological relationships than simply ranking features. We demonstrate Foresight on both an artificial dataset and a large gene expression dataset of breast cancer patients. Using the latter dataset we show that our approach retrieves biologically relevant features and provides a rich description of the interactions and correlation structure between these features.

bioinformatics↗

bioRxiv2017

Phylostratigraphic analysis of tumor and developmental transcriptomes reveals relationship between oncogenesis, phylogenesis and ontogenesis

The question of the existence of cancer is inadequately answered by invoking somatic mutations or the disruptions of cellular and tissue control mechanisms. As such uniformly random events alone cannot account for the almost inevitable occurrence of an extremely complex process such as cancer. In the different epistemic realm, an ultimate explanation of cancer is that cancer is a reversion of a cell to an ancestral pre-Metazoan state, i.e. a cellular form of atavism. Several studies have suggested that genes involved in cancer have evolved at particular evolutionary time linked to the unicellular-multicellular transition. Here we used a refined phylostratigraphic analysis of evolutionary ages of the known genes/pathways associated with cancer and the genes differentially expressed between normal and cancer tissue as well as between embryonic and mature (differentiated) cells. We found that cancer-specific transcriptomes and cancer-related pathways were enriched for genes that evolved in the pre-Metazoan era and depleted of genes that evolved in the post-Metazoan era. By contrast an opposite relation was found for cell maturation: the age distribution frequency of the genes expressed in differentiated epithelial cells were enriched for post-Metazoan genes and depleted of pre-Metazoan ones. These findings support the atavism theory that cancer cells manifest the reactivation of an ancient ancestral state featuring unicellular modalities. Thus our bioinformatics analyses suggest that not only does oncogenesis recapitulate ontogenesis, and ontogenesis recapitulates phylogenesis, but also oncogenesis recapitulates phylogenesis. This more encompassing perspective may offer a natural organizing framework for genetic alterations in cancers and point to new treatment options that target the genes controlling the atavism transition.\n\nOne Sentence SummaryTracing cancer gene evolutionary ages revealed that cancer reverts to a pre-existing early Metazoan state.

cancer biology↗

bioRxiv2017

Extracting Intercellular Signaling Network Of Cancer Tissues Using Ligand-Receptor Expression Patterns From Whole-Tumor And Single-Cell Transcriptomes

Many behaviors of cancer, such as progression, metastasis and drug resistance etc., cannot be fully understood by genetic mutations or intracellular signaling alone. Instead, they are emergent properties of cell community that forms a tumor. Studies of tumor heterogeneity reveal that many cancer behaviors critically depend on the intercellular communication mediated by secreted signaling ligands and their cognate receptors that take place between cancer cells and stromal cells. Owing to systematic cancer omics efforts, we studied such cell-cell interactions using data from cancer transcriptome database. We curated a list of more than 2,500 ligand-receptor pairs and developed a method to identify their enrichment in tumors from TCGA pancancer data and to build a cell interaction network from single-cell data for the case of melanoma. Using the specificity of the ligand-receptor interaction and their expressions measured in individual cells, we built a map of a cell-cell communication network which indicate what signal is exchanged between which cell types. Such networks establish a new formal phenotypes which are embodied by the cell communication structure - it may offer new opportunities to identify the molecular signatures which may influence cancer cell behaviors by changing cell population dynamics.

systems biology↗