What is the main focus of your group?
At the Institute for Computational Biomedicine, we develop and apply computational methods to analyse complex biomedical data. Our goal is to acquire a functional understanding of molecular mechanisms in health and disease. We apply these insights to improve diagnosis and to develop novel therapeutics that we apply across diverse disease areas, including oncology, cardiology, nephrology, neurology, and infectious diseases.
We work closely with experimental and clinical groups within Heidelberg University’s Medical Faculty and University Hospital, as well as national and international collaborators. We are connected with the EMBL via the Molecular Medicine Partnership Unit on Chronic Kidney Disease, and we are also part of ELLIS, the European initiative for Artificial Intelligence.
A key emphasis is to build models that are both mechanistic (to provide understanding), predictive and hypothesis-generating, using methods from statistics, machine learning, and engineering. To build these models, we combine existing biological knowledge with large-scale molecular data, in particular genomics, transcriptomics and proteomics, obtained from patients or appropriate model systems.
How has participating in TransQST impacted your institution?
TransQST brought together an excellent team of leading scientists from academia and industry. The good balance between experimental and computational groups led to fruitful collaborations and to the development of new computational tools which can be used both by the consortium partners and other researchers within our organisation.
What challenges has UKHD faced across the project?
One of the main challenges in TransQST was for us to develop methods tailored to the datasets used in toxicology. Our computational research group had a strong background in modelling perturbational phosphoproteomics datasets, whereas in toxicology, the most abundant datasets are unperturbed transcriptomics.
Did you develop new computational tools for toxicology studies?
Our tools focused on modelling signalling datasets, however, transcriptomic datasets are more abundant in the field of toxicology studies. To be able to infer the signalling network from transcriptomics datasets, we developed the CARNIVAL tool. CARNIVAL generates hypotheses on potential upstream alterations that propagate through signalling networks, providing insights into mechanisms of drug action. CARNIVAL integrates different sources of prior knowledge including signed and directed protein–protein interactions, transcription factor targets, and pathway signatures. The use of prior knowledge in CARNIVAL enables capturing a broad set of upstream cellular processes and regulators, leading to a higher accuracy when benchmarked against related tools.
In collaboration with Maastricht University, we applied CARNIVAL to infer upstream signalling networks deregulated in drug-induced liver injury (DILI) from gene expression data from publicly available databases (openTG-GATEs). We focused on six compounds that induce observable histopathologies linked to DILI from repeated dosing experiments in rats and compared responses in vitro and in vivo to identify potential cross-platform concordances in rats as well as network similarities between rat and human, demonstrating shared pathways and network motifs between compounds. These pathways and motifs resulted in similar pathologies in rats, but not in humans. In particular, the causal interactions “LCK activates SOCS3, which in turn inhibits TFDP1” was commonly identified as a regulatory path among the fibrosis-inducing compounds. This potential pathology-inducing circuit illustrates the value of our approach to generate hypotheses that can be further validated experimentally.
How can the consortium partners and scientists beyond the consortium use your tools?
We and the entire TransQST consortium believe in open science and are dedicated to open-source tools, therefore all of our computational analysis pipelines are available for the public on our github, www.github.com/saezlab. Further, we disseminate our tools by publishing them in well-known repositories, such as Bioconductor (www.bioconductor.org).