The integration of single cell and spatial transcriptomics provides a new approach to profile human disease pathology in situ. Here, I will introduce our work on dissecting lung alveolar damage in severe COVID-19 using a new single cell atlas and transcriptome wide spatial profiling of post-mortem lung tissue. First, we generated a comprehensive single-cell lung cell atlas through integration of multiple healthy and COVID-19 datasets. Second, we generated a spatially resolved transcriptomic dataset of diffuse alveolar damage (DAD) across different stages of pathology using the Nanostring WTA technology. To resolve changes in cell type abundance across progressive pathology, we integrated our single cell and spatial transcriptomic datasets. We identified dynamic sets of immune and stromal cells and tissue microenvironments that distinguish early (exudative) and late (organised) alveolar damage. Finally, we could re-map pathological phenotypes in our single-cell transcriptomic reference using pathology biomarkers identified from spatial data. Our work identifies candidate molecular and cellular targets of novel therapies for COVID-19 in the respiratory system.

TBD

We develop artificial intelligence (AI) methods for extracting new biomedical knowledge from the wiring patterns of systems-level, heterogeneous omics data. Our graphlet-based and other methods uncover the patterns in molecular (omics) networks and in the multi-scale organization of these networks indicative of biological function, translating the information hidden in the network topology into domain-specific knowledge. Also, we introduce a versatile data fusion (integration) machine learning (ML) framework to address key challenges in precision medicine from the wiring patterns of omics network data: better stratification of patients, prediction of driver genes in cancer, and re-purposing of approved drugs to particular patients and patient groups, including Covid-19 patients. Our new methods stem from novel network science algorithms coupled with graph-regularized non-negative matrix tri-factorization (NMTF), a machine learning technique for dimensionality reduction, inference and co-clustering of heterogeneous datasets. We utilize our new framework to develop methodologies for understanding the molecular organization of the omics data embedding space.

I will discuss a unifying statistical formulation for many fundamental problems in genome science and develop a reference-free, highly efficient algorithm that solves it.  This formulation allows us to construct an algorithm that performs inference on raw reads, avoiding references completely. We illustrate the power of our approach for new data-driven biological discovery with examples of novel single-cell resolved, cell-type-specific isoform expression, including splicing, expression in the major histocompatibility complex, and de novo prediction of viral protein adaptation including in SARS-CoV-2.

Complex traits are established through the joint influences of multiple genetic and environmental perturbations. There is a shortage of generalizable principles explaining how molecular networks integrate genetic and environmental effects ultimately leading to complex cellular and organismal traits. In particular, it is poorly understood when and how genetic perturbations lead to molecular changes that are confined to small parts of a network versus when they lead to large-scale adaptations of global network states. Here, we present a concept classifying genetic effects as local, regional or global depending on what fraction of a molecular network they affect. We exemplify this notion using transcriptome, proteome and phospho-proteome profiling of genetically heterogeneous populations of yeast strains, which we integrate with an array of cellular traits. Our analysis identified a central gauge of the yeast molecular network that is related to PKA and TOR (PT) signaling. The resulting ‘PT state’ could be summarized in a single value that explained large parts of the molecular configuration of the strains. This PT state associated with a specific balance between cellular processes spanning energy- and amino acid metabolism, transcription, translation, cell cycle control and cellular stress response. Carbon source quality, oxidative stress, and gene-environment interactions caused monotonic shifts of the molecular network state along the same axis. We further show that complex traits like heat stress resistance and longevity (stationary phase viability) result from the synthesis of genetic effects modulating this PT state with global network effects, plus much more trait-specific effects modulating only small parts of the network. Our work provides a rational for the conditions under which genetic effects propagate through molecular networks with pleiotropic consequences.

TBD

Approaches for the identification of disease causal mutations are widely applied in research and clinical settings, but interpretation and ranking of the resulting variants remains challenging. Combined Annotation Dependent Depletion (CADD, https://cadd-sv.bihealth.org/) integrates annotations by contrasting variants that survived purifying selection along the human lineage with simulated mutations to score short sequence variants (SNVs, InDels, multi-allelic substitutions). Since its publication (Kircher, Witten et al. Nat Genet. 2014), CADD was well adopted by the community and minor adjustments and fixes were released since, including the native support of both GRCh37 and GRCh38 assemblies (Rentzsch et al. NAR 2019). Recently, we assessed existing deep neural network (DNN) models for splice effects with the Multiplexed Functional Assay of Splicing using Sort-seq dataset (MFASS, Cheung et al. Mol Cell. 2019). We selected two DNN models based only on genomic sequence, MMSplice and SpliceAI, which showed the best performance for integration into CADD (Rentzsch et al. Genome Med. 2021). The DNN scores boosted CADD's predictions for splice effects and we noted that while the DNN scores have superior performance on splice variants, they fail to account for nonsense and missense effects of the same variants. This suggests that variant prioritization will improve with more domain-specific information and underlines the importance of identifying additional such features, e.g. for regulatory sequences. With rapid advances in the identification of structural variants (SVs), we decided to apply the general concept of CADD to score them (CADD-SV, https://cadd-sv.bihealth.org/). While methods utilizing individual mechanistic principles like the deletion of coding sequence or 3D architecture disruptions were available, a comprehensive tool that uses the broad spectrum of available SV annotations was missing. We show that CADD-SV scores are predictive of pathogenicity and population frequency and that CADD-SV's ability to prioritize pathogenic variants exceeds that of existing methods like SVScore and AnnotSV (Kleinert & Kircher, Genome Res. 2022). Our results highlight advantages of the CADD approach, like profiting from a large training data set covering diverse and rare feature annotations without major ascertainment effects from historic and on-going variant collections.

I will describe two projects that aim to better dissect the causal chain from functional genetic variant through molecular intermediates and finally to organismal trait or disease risk. In the first, we are using pooled profiling of RNA binding protein (RBPs, splice factors) binding across individuals to measure and then computationally model genetic effects on both binding and RNA splicing. In the second, we have developed a causal network inference method that scales to hundreds of nodes by leveraging convex optimization.