Robust evaluation of deep learning-based representation methods for survival and gene essentiality prediction on bulk RNA-seq data.

Deep learning (DL) has shown potential to provide powerful representations of bulk RNA-seq data in cancer research. However, there is no consensus regarding the impact of design choices of DL approaches on the performance of the learned representation, including the model architecture, the training methodology and the various hyperparameters. To address this problem, we evaluate the performance of various design choices of DL representation learning methods using TCGA and DepMap pan-cancer datasets and assess their predictive power for survival and gene essentiality predictions. We demonstrate that baseline methods achieve comparable or superior performance compared to more complex models on survival predictions tasks. DL representation methods, however, are the most efficient to predict the gene essentiality of cell lines. We show that auto-encoders (AE) are consistently improved by techniques such as masking and multi-head training. Our results suggest that the impact of DL representations and of pretraining are highly task- and architecture-dependent, highlighting the need for adopting rigorous evaluation guidelines. These guidelines for robust evaluation are implemented in a pipeline made available to the research community.

Learning interpretable causal networks from very large datasets, application to 400,000 medical records of breast cancer patients.

Discovering causal effects is at the core of scientific investigation but remains challenging when only observational data are available. In practice, causal networks are difficult to learn and interpret, and limited to relatively small datasets. We report a more reliable and scalable causal discovery method (iMIIC), based on a general mutual information supremum principle, which greatly improves the precision of inferred causal relations while distinguishing genuine causes from putative and latent causal effects. We showcase iMIIC on synthetic and real-world healthcare data from 396,179 breast cancer patients from the US Surveillance, Epidemiology, and End Results program. More than 90% of predicted causal effects appear correct, while the remaining unexpected direct and indirect causal effects can be interpreted in terms of diagnostic procedures, therapeutic timing, patient preference or socio-economic disparity. iMIIC's unique capabilities open up new avenues to discover reliable and interpretable causal networks across a range of research fields.

Interactive exploration of a global clinical network from a large breast cancer cohort.

Despite unprecedented amount of information now available in medical records, health data remain underexploited due to their heterogeneity and complexity. Simple charts and hypothesis-driven statistics can no longer apprehend the content of information-rich clinical data. There is, therefore, a clear need for powerful interactive visualization tools enabling medical practitioners to perceive the patterns and insights gained by state-of-the-art machine learning algorithms. Here, we report an interactive graphical interface for use as the front end of a machine learning causal inference server (MIIC), to facilitate the visualization and comprehension by clinicians of relationships between clinically relevant variables. The widespread use of such tools, facilitating the interactive exploration of datasets, is crucial both for data visualization and for the generation of research hypotheses. We demonstrate the utility of the MIIC interactive interface, by exploring the clinical network of a large cohort of breast cancer patients treated with neoadjuvant chemotherapy (NAC). This example highlights, in particular, the direct and indirect links between post-NAC clinical responses and patient survival. The MIIC interactive graphical interface has the potential to help clinicians identify actionable nodes and edges in clinical networks, thereby ultimately improving the patient care pathway.

Inferring Gene Networks in Bone Marrow Hematopoietic Stem Cell-Supporting Stromal Niche Populations.

The cardinal property of bone marrow (BM) stromal cells is their capacity to contribute to hematopoietic stem cell (HSC) niches by providing mediators assisting HSC functions. In this study we first contrasted transcriptomes of stromal cells at different developmental stages and then included large number of HSC-supportive and non-supportive samples. Application of a combination of algorithms, comprising one identifying reliable paths and potential causative relationships in complex systems, revealed gene networks characteristic of the BM stromal HSC-supportive capacity and of defined niche populations of perivascular cells, osteoblasts, and mesenchymal stromal cells. Inclusion of single-cell transcriptomes enabled establishing for the perivascular cell subset a partially oriented graph of direct gene-to-gene interactions. As proof of concept we showed that R-spondin-2, expressed by the perivascular subset, synergized with Kit ligand to amplify ex vivo hematopoietic precursors. This study by identifying classifiers and hubs constitutes a resource to unravel candidate BM stromal mediators.