In this talk, I will present current activities of the Computational Systems Biology group at IBM Research, Zurich, focused on the inference and exploitation of networks of molecular interactions. Focusing first on the problem of network inference, a long-standing challenge for which many methods have been proposed, I will discuss how no single inference method performs optimally across all data sets.
However, a Wisdom of the Crowds approach based on the integration of multiple inference methods can increase the robustness and high performance of the inferred networks. To that aim, we have developed COSIFER, a web-based platform that enables the inference of molecular networks using different approaches and consensus strategies. Next, I will introduce INtERAcT, an approach to extract information about molecular interactions from a text corpus in a completely unsupervised manner. INtERAcT exploits word embeddings, a state-of-the-art technology for language modelling based on deep learning that does not require text labeling for training or domain-specific knowledge, and hence, can be easily applied to different scientific domains.
Moving into the applications, I will explain how prior information about the molecular interactions in a cell can be encoded in a network, which can be further used for gene prioritization. Such strategy is exploited by NetBiTE with the goal of identifying anti-cancer drug sensitivity biomarkers. Finally, I will discuss how a probabilistic application of network dynamics can enable the reconstruction of the cell-signaling dynamics using single-cell omics.[-]

There is an emerging consensus in the transdiciplinary literature that the ultimate goal of regression analysis is to model the conditional distribution of an outcome, given a set of explanatory variables or covariates. This new approach is called "distributional regression", and marks a clear break from the classical view of regression, which has focused on estimating a conditional mean or quantile only. Isotonic Distributional Regression (IDR) learns conditional distributions that are simultaneously optimal relative to comprehensive classes of relevant loss functions, subject to monotonicity constraints in terms of a partial order on the covariate space. This IDR solution is exactly computable and does not require approximations nor implementation choices, except for the selection of the partial order. Despite being an entirely generic technique, IDR is strongly competitive with state-of-the-art methods in a case study on probabilistic precipitation forecasts from a leading numerical weather prediction model.

Joint work with Alexander Henzi and Johanna F. Ziegel.[-]