Connectivity Homology Enables Inter-Species Network Models of Synthetic Lethality

Abstract

Synthetic lethality is a genetic interaction wherein two otherwise nonessential genes cause cellular inviability when knocked out simultaneously. Drugs can mimic genetic knock-out effects; therefore, our understanding of promiscuous drugs, polypharmacology-related adverse drug reactions, and multi-drug therapies, especially cancer combination therapy, may be informed by a deeper understanding of synthetic lethality. However, the colossal experimental burden in humans necessitates in silico methods to guide the identification of synthetic lethal pairs. Here, we present SINaTRA (Species-INdependent TRAnslation), a network-based methodology that discovers genome-wide synthetic lethality in translation between species. SINaTRA uses connectivity homology, defined as biological connectivity patterns that persist across species, to identify synthetic lethal pairs. Importantly, our approach does not rely on genetic homology or structural and functional similarity, and it significantly outperforms models utilizing these data. We validate SINaTRA by predicting synthetic lethality in S. pombe using S. cerevisiae data, then identify over one million putative human synthetic lethal pairs to guide experimental approaches. We highlight the translational applications of our algorithm for drug discovery by identifying clusters of genes significantly enriched for single- and multi-drug cancer therapies.

Biological network topology features predict gene dependencies in cancer cell lines

“…To improve performance in cross cell line classification each cell line's feature set was normalised (Jacunski et al, 2015). To ensure unbiased validation we held-out 20% of this data to be used as a test set leaving 80% to be used as training data.…”

A Multi-Species Functional Embedding Integrating Sequence and Network Structure

“…synthetic lethality). More recently, Jacunski, et al [26] introduced a general purpose cross-species node vector representation, but their work is focused on training classifiers using the vectors as features. Although they show good performance on these prediction tasks, their representations are derived from summary statistics, and we show the distances in their vector space are not strongly correlated with functional similarity.…”

“…In particular, we compare Handl to two standard network alignment algorithms, IsoRank and HubAlign [36,24], from which it is possible to extract cross-species protein metrics that can be interpreted as similarity scores, and thus serve as a natural comparison point for HANDL homology scores. We also compare Handl to the sequence alignment algorithm BLAST [3] and SINaTRA [26]. We then define a pair of proteins (p i , p j ) from two species to be k-functionally similar if both p i and p j are annotated by the same GO term and, in each species, that GO term is associated with at most k proteins (see Section 4.4 for details on processing of GO).…”

“…We then define a pair of proteins (p i , p j ) from two species to be k-functionally similar if both p i and p j are annotated by the same GO term and, in each species, that GO term is associated with at most k proteins (see Section 4.4 for details on processing of GO). This metric for functional similarity was also used (without naming it) in [26] and is similar to the Gene Ontology Consistency (GOC), or Functional Consistency (FC) metrics that are also used to evaluate network alignment algorithms [12,24].…”

“…The dominant paradigm for this work is based on matching nodes; in some cases, this is done using cross-species node similarity scores in an intermediate step (e.g., [58,36,39,37,44,24]). Working in another direction, Jacunski, et al [26] introduced the notion of connectivity homology to relate genes in different species by their position in corresponding PPI networks. While not the focus of their paper, they show that pairs of genes with the same or similar functions have lower connectivity homology dissimilarity scores.…”