Estimation of the proteomic cancer co-expression sub networks by using association estimators

k- Strong Inference Algorithm: A Hybrid Information Theory Based Gene Network Inference Algorithm

Gene networks allow researchers to understand the underlying mechanisms between diseases and genes while reducing the need for wet lab experiments. Numerous gene network inference (GNI) algorithms have been presented in the literature to infer accurate gene networks. We proposed a hybrid GNI algorithm, k-Strong Inference Algorithm (ksia), to infer more reliable and robust gene networks from omics datasets. To increase reliability, ksia integrates Pearson correlation coefficient (PCC) and Spearman rank correlation coefficient (SCC) scores to determine mutual information scores between molecules to increase diversity of relation predictions. To infer a more robust gene network, ksia applies three different elimination steps to remove redundant and spurious relations between genes. The performance of ksia was evaluated on microbe microarrays database in the overlap analysis with other GNI algorithms, namely ARACNE, C3NET, CLR, and MRNET. Ksia inferred less number of relations due to its strict elimination steps. However, ksia generally performed better on Escherichia coli (E.coli) and Saccharomyces cerevisiae (yeast) gene expression datasets due to F- measure and precision values. The integration of association estimator scores and three elimination stages slightly increases the performance of ksia based gene networks. Users can access ksia R package and user manual of package via https://github.com/ozgurcingiz/ksia .

A mutual information-based in vivo monitoring of adaptive response to targeted therapies in melanoma

•

In vivo dependencies should be quantified in adaptive resistance mechanistic studies

•

Mutual information (MI) quantifies any type rather than just linear dependencies

•

MI outperforms classic expression correlation coefficients

•

Adaptive response to small-molecules inhibitors can be monitored in vivo using MI

•

Strategies that prevent adaptive resistance can be monitored in vivo using MI

The mechanisms of adaptive resistance to genetic-based targeted therapies of solid malignancies have been the subject of intense research. These studies hold great promise for finding co-targetable hub/pathways which in turn would control the downstream non-genetic mechanisms of adaptive resistance. Many such mechanisms have been described in the paradigmatic BRAF-mutated melanoma model of adaptive response to BRAF inhibition. Currently, a major challenge for these mechanistic studies is to confirm in vivo, at the single-cell proteomic level, the existence of dependencies between the co-targeted hub/pathways and their downstream effectors. Moreover, the drug-induced in vivo modulation of these dependencies needs to be demonstrated. Here, we implement such single-cell-based in vivo expression dependency quantification using immunohistochemistry (IHC)-based analyses of sequential biopsies in two xenograft models. These mimic phase 2 and 3 trials in our own therapeutic strategy to prevent the adaptive response to BRAF inhibition. In this mechanistic model, the dependencies between the targeted Li₂CO₃-inducible hub HuR and the resistance effectors are more likely time-shifted and transient since the minority of HuR^Low cells, which act as a reservoir of adaptive plasticity, switch to a HuR^High state as they paradoxically proliferate under BRAF inhibition. Nevertheless, we show that a copula/kernel density estimator (KDE)-based quantification of mutual information (MI) efficiently captures, at the individual level, the dependencies between HuR and two relevant resistance markers pERK and EGFR, and outperforms classic expression correlation coefficients. Ultimately, the validation of MI as a predictive IHC-based metric of response to our therapeutic strategy will be carried in clinical trials.