Search

Abstract Background There is an increasing interest in the use of Deep Learning (DL) based methods as a supporting analytical framework in oncology. However, most direct applications of DL will deliver models with limited transparency and explainability, which constrain their deployment in biomedical settings. Methods This systematic review discusses DL models used to support inference in cancer biology with a particular emphasis on multi-omics analysis. It focuses on how existing models address the need for better dialogue with prior knowledge, biological plausibility and interpretability, fundamental properties in the biomedical domain. For this, we retrieved and analyzed 42 studies focusing on emerging architectural and methodological advances, the encoding of biological domain knowledge and the integration of explainability methods. Results We discuss the recent evolutionary arch of DL models in the direction of integrating prior biological relational and network knowledge to support better generalisation (e.g. pathways or Protein-Protein-Interaction networks) and interpretability. This represents a fundamental functional shift towards models which can integrate mechanistic and statistical inference aspects. We introduce a concept of bio-centric interpretability and according to its taxonomy, we discuss representational methodologies for the integration of domain prior knowledge in such models. Conclusions The paper provides a critical outlook into contemporary methods for explainability and interpretability used in DL for cancer. The analysis points in the direction of a convergence between encoding prior knowledge and improved interpretability. We introduce bio-centric interpretability which is an important step towards formalisation of biological interpretability of DL models and developing methods that are less problem- or application-specific.

Abstract Background Spatial interactions and insulation of chromatin regions are associated with transcriptional regulation. Domains of frequent chromatin contacts are proposed as functional units, favoring and delimiting gene regulatory interactions. However, contrasting evidence supports the association between chromatin domains and transcription. Result Here, we assess gene co-regulation in chromatin domains across multiple human cancers, which exhibit great transcriptional heterogeneity. Across all datasets, gene co-regulation is observed only within a small yet significant number of chromatin domains. We design an algorithmic approach to identify differentially active domains (DADo) between two conditions and show that these provide complementary information to differentially expressed genes. Domains comprising co-regulated genes are enriched in the less active B sub-compartments and for genes with similar function. Notably, differential activation of chromatin domains is not associated with major changes of domain boundaries, but rather with changes of sub-compartments and intra-domain contacts. Conclusion Overall, gene co-regulation is observed only in a minority of chromatin domains, whose systematic identification will help unravel the relationship between chromatin structure and transcription.

Abstract Background Chromatin folding gives rise to structural elements among which are clusters of densely interacting DNA regions termed topologically associating domains (TADs). TADs have been characterized across multiple species, tissue types, and differentiation stages, sometimes in association with regulation of biological functions. The reliability and reproducibility of these findings are intrinsically related with the correct identification of these domains from high-throughput chromatin conformation capture (Hi-C) experiments. Results Here, we test and compare 22 computational methods to identify TADs across 20 different conditions. We find that TAD sizes and numbers vary significantly among callers and data resolutions, challenging the definition of an average TAD size, but strengthening the hypothesis that TADs are hierarchically organized domains, rather than disjoint structural elements. Performances of these methods differ based on data resolution and normalization strategy, but a core set of TAD callers consistently retrieve reproducible domains, even at low sequencing depths, that are enriched for TAD-associated biological features. Conclusions This study provides a reference for the analysis of chromatin domains from Hi-C experiments and useful guidelines for choosing a suitable approach based on the experimental design, available data, and biological question of interest.