Mapping the multiscale structure of biological systems

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

It is a challenge in single-cell analyses to deconvolve the relationship between cell type and time, resulting in the development of a diversity of algorithms to infer the trajectory of cell differentiation across time.

“it is a challenge in single-cell analyses to deconvolve the relationship between cell type and time, resulting in the development of a diversity of algorithms to infer the trajectory of cell differentiation across time (Saelens et al., 2019).”

• cites Saelens et al., 2019 sourceNot checked yet

  Not checked in the free run.

May combine two claims.

Hi-C records pairs of genomic loci at close proximity, and certain pairs emerge close in three dimensions although separated at large nucleotide distances along the linear chromosome.

“Moving from proteins to DNA, the family of techniques known as chromosome conformation capture, including 3C (Dekker et al., 2002) and Hi-C (Lieberman-Aiden et al., 2009), record pairs of genomic loci that are at close proximity.”

• cites Lieberman-Aiden et al., 2009 sourceNot checked yet

  Not checked in the free run.

May combine two claims.

Genetic interactions can be observed among genes participating in distinct parallel biological pathways, so genes with similar genetic interaction profiles do not always encode proteins in close physical proximity.

“Although genes with similar genetic interaction profiles can encode proteins in close physical proximity, this does not always have to be the case, as genetic interactions can also be observed among genes participating in distinct parallel biological pathways (Gonatopoulos-Pournatzis et al., 2020).”

• cites Gonatopoulos-Pournatzis et al., 2020 sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

The community detection approach was well illustrated in recent analyses of protein-protein interaction networks in yeast and human.

“The community detection approach was well illustrated in recent analyses of protein-protein interaction networks in yeast (Dutkowski et al., 2013) and human (Huttlin et al., 2015).”

• cites Dutkowski et al., 2013 sourceNot checked yet

  Not checked in the free run.

• cites Huttlin et al., 2015 sourceNot checked yet

  Not checked in the free run.

May combine two claims.

Amino-acid proximity networks are central to many efforts to model protein structure and to protein structure prediction.

“Such amino-acid proximity networks, whether experimentally generated or inferred, are central to many efforts to model protein structure. They are also central to protein structure prediction.”

• cites Senior et al., 2019 sourceNot checked yet

  Not checked in the free run.

• cites Senior et al., 2020 sourceNot checked yet

  Not checked in the free run.

May combine two claims.

Cell-cell similarity networks from single-cell RNA sequencing can be analyzed to track cell developmental trajectories over time or to recognize groups of cells forming a distinct cell type or tissue.

“The resulting cell-cell similarity network can then be analyzed to track cell developmental trajectories over time (Farrell et al., 2018; Wagner et al., 2018; Zhong et al., 2018) or to recognize groups of cells forming a distinct cell type or tissue (Jaitin et al., 2014; Zeisel et al., 2015).”

• cites Farrell et al., 2018 sourceNot checked yet

  Not checked in the free run.

• cites Wagner et al., 2018 sourceNot checked yet

  Not checked in the free run.

• cites Zeisel et al., 2015 sourceNot checked yet

  Not checked in the free run.

Classified as probably not a substantial claim. May combine two claims.

Developing a multiscale representation requires analyzing communities in biological networks at more than one scale, defining and identifying communities at each scale, and tracing containment to infer a hierarchical map.

“Developing a multiscale representation requires analyzing communities in biological networks at more than one scale. Communities (clusters of highly interconnected nodes) are first defined and identified at each of multiple scales; subsequently, communities that are contained within other communities, in whole or in part, can be traced to infer a hierarchical map (Figure 2) (Dotan-Cohen et al., 20”

• cites Dotan-Cohen et al., 2009 sourceNot checked yet

  Not checked in the free run.

• cites Jaimovich et al., 2010 sourceNot checked yet

  Not checked in the free run.

• cites Ravasz et al., 2002 sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

The Mashup algorithm analyzes the topology of each network separately and learns a single weighted network representation that best explains the topological patterns observed across all networks.

“In a related method, the Mashup algorithm analyzes the topology of each network separately and learns a single weighted network representation that best explains the topological patterns observed across all networks (Cho et al., 2016).”

• cites Cho et al., 2016 sourceNot checked yet

  Not checked in the free run.

Classified as probably not a substantial claim.

Once multiple datasets have been combined into a unified network, community detection algorithms may be run on this network to build hierarchical maps.

“Once the data have been combined into a unified network, community detection algorithms may then be run on this network to build hierarchical maps as described above.”

Probably not a substantial claim: may not pass S4 (based).

At any level of the biological hierarchy, an ever-increasing diversity of technologies can characterize biological units and their relations, yielding large networks of physical or functional proximities.

“At any level of the hierarchy, an ever-increasing diversity of technologies can be applied to characterize the corresponding biological units and their relations, resulting in large networks of physical or functional proximities”

May combine two claims.

In the human protein interaction study, clique percolation determined high-level protein communities, and modularity-based clustering subdivided them into more granular subsystems, resulting in a multi-layer hierarchical structure.

“In the human study, the authors first performed clique percolation to determine high-level protein communities; subsequently, they ran modularity-based clustering to further subdivide these communities into more granular subsystems, resulting in a multi-layer hierarchical structure (Figure 3A).”

• cites Huttlin et al., 2015 sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S2 (checkable).

Anatomical dissection results can be thought of as generating a network of tissue types interrelated by physical or functional proximity.

“Finally, at the highest level in the structural hierarchy of an organism, results from anatomical dissection can be thought of as generating a network of tissue types interrelated to one another by physical or functional proximity.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

There is an important tradeoff between conveying the complete information represented in the hierarchy and avoiding a graphic that is too complex to interpret or investigate.

“In this common case, there is an important tradeoff between conveying the complete information represented in the hierarchy while avoiding a graphic that is too complex to interpret or investigate.”

May combine two claims.

In the DCell study, the structure of a deep neural network was constrained to mirror the hierarchy of cellular components recorded by GO or a multiscale map of cellular components derived from molecular interaction data, and DCell learned to translate genotype to phenotype and capture the functional state of cell subsystems throughout the hierarchy.

“In this study, the structure of a deep neural network (i.e. configuration of artificial neurons) was constrained to mirror the hierarchy of cellular components recorded by GO or, alternatively, a multiscale map of cellular components derived from molecular interaction data. This model, called DCell, was able to learn not only to translate genotype to phenotype, but also to capture the functional”

• cites Ma et al., 2018Not checked yet

  Not checked in the free run.

May combine two claims.

Even with a multiscale model, translation of genotype to phenotype can be complex, and the use of standard black-box machine learning algorithms prevents straightforward biological interpretation.

“Even with a multiscale model, the translation of genotype to phenotype can be complex, and the use of standard “black-box” machine learning algorithms prevents a straightforward biological interpretation (Rudin, 2019).”

• cites Rudin, 2019 sourceNot checked yet

  Not checked in the free run.

May combine two claims.

Gaudelet et al. modeled a neural network structure on the multiscale functional organization of a cell by integrating gene-pathway annotations in the Reactome database, and their model accurately determined patient diagnoses based on gene expression data.

“In a related study, Gaudelet et al. modeled a neural network structure on the multiscale functional organization of a cell by integrating gene-pathway annotations in the Reactome database; their model was able to accurately determine patient diagnoses based on gene expression data.”

• cites Gaudelet et al., 2020 sourceNot checked yet

  Not checked in the free run.

May combine two claims.

Gaudelet et al. modeled pairwise protein interactions, complexes, and pathways as multiscale hypernetworks, and analysis of these hypergraphs resulted in determining protein function for many uncharacterized genes.

“For example, Gaudelet et al. modeled pairwise protein interactions, complexes, and pathways as multiscale hypernetworks; analysis of these hypergraphs resulted in determining protein function for many uncharacterized genes (Gaudelet et al., 2018).”

• cites Gaudelet et al., 2018 sourceNot checked yet

  Not checked in the free run.

May combine two claims.

The Ohmnet multiscale approach at the level of biological tissues enabled prediction of tissue-specific cellular functions for different proteins, outperforming competing methods that did not account for the multiscale hierarchical organization of tissues.

“This multiscale approach at the level of biological tissues enabled prediction of tissue-specific cellular functions for different proteins, outperforming competing methods which did not account for the multiscale hierarchical organization of tissues.”

• cites Zitnik and Leskovec, 2017Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

Ontology alignment seeks to address the challenge of determining entities in different hierarchies that represent the same concept, including how to handle multiple overlapping entities and differences in ontology organization.

“For this purpose, the area known as ”ontology alignment” seeks to address the challenge of determining entities in different hierarchies that represent the same concept, including how to handle multiple overlapping entities and differences in ontology organization.”

Classified as probably not a substantial claim.

Visualization of results as a hierarchy communicates the hierarchical relationships among subsystems across multiple levels.

“While many applications of clustering and community detection show the detected communities as a flat list or network (Figure 3A), visualization of results as a hierarchy (Figure 5A) communicates the hierarchical relationships among subsystems across multiple levels.”

May combine two claims.

Yu et al. proposed an explicit hierarchical approach for translation of genotype to phenotype in which mutations were propagated upward through a hierarchy of cellular components and processes, and mutation states of entities were used as features to explain phenotype.

“An explicit hierarchical approach for translation of genotype to phenotype was proposed by (Yu et al., 2016), in which mutations were propagated upward through a hierarchy of cellular components and processes. The mutation states of these entities (i.e. entities containing mutated genes) were then used as features to explain phenotype, enabling the appropriate scale in the cell to be used for”

• cites Yu et al., 2016Not checked yet

  Not checked in the free run.

May combine two claims.

Biological ontologies are large hierarchies with a DAG structure, so existing tools for visualizing ontologies can have direct application to data-driven multiscale models.

“Biological ontologies are large hierarchies which also have a DAG structure; as such, existing tools for visualizing ontologies can also have direct application to data-driven multiscale models (Binns et al., 2009; Eden et al., 2009; Sealfon et al., 2006; Supek et al., 2011; Zhu et al., 2019).”

• cites Binns et al., 2009 sourceNot checked yet

  Not checked in the free run.

• cites Eden et al., 2009 sourceNot checked yet

  Not checked in the free run.

• cites Sealfon et al., 2006 sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

Multiscale models have great potential for understanding function and predicting outcomes in biological systems, since knowledge of higher-order protein assemblies in which proteins participate provides evidence for their function and relevance to diseases.

“Multiscale models have great potential for the applications of understanding function and predicting outcomes in biological systems, since knowledge of the higher-order protein assemblies in which proteins participate provides evidence for their function and relevance to particular diseases.”

May combine two claims.

Another study generated a multiscale hierarchical map from a co-expression network of a pathogenic fungus; this map was used to identify clusters in the hierarchy associated with plant cell infection, determining genes not previously reported to function in these pathways.

“Another study generated a multiscale hierarchical map from a co-expression network of a pathogenic fungus; this map was used to identify clusters in the hierarchy associated with plant cell infection, determining genes not previously reported to function in these pathways (Ames, 2017).”

• cites Ames, 2017 sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

Systems not found in GO may correspond to novel discoveries.

“Conversely, systems not found in GO may correspond to novel discoveries.”

May combine two claims.

DrugCell is an interpretable deep learning model of cancer cell proliferation trained on the response of tumor cell lines to a large panel of drugs, combining genotype and drug structure to predict drug response along with specific cellular subsystems mediating that response.

“This thread was recently extended through the development of DrugCell, an interpretable deep learning model of cancer cell proliferation (Kuenzi et al., 2020). Here, the model was trained on the response of tumor cell lines to a large panel of drugs, combining genotype and drug structure to predict the drug response along with the specific cellular subsystems mediating that response.”

• cites Kuenzi et al., 2020 sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Neural networks have been used to perform interpretable deep learning on biological networks to differentiate cell types and determine regulatory differences in single-cell RNA sequencing data.

“Related work has used neural networks to perform interpretable deep learning on biological networks to differentiate cell types and determine regulatory differences in single cell RNA sequencing data (Fortelny and Bock, 2020).”

• cites Fortelny and Bock, 2020 sourceNot checked yet

  Not checked in the free run.

May combine two claims.

The Yu et al. hierarchical approach accurately predicted yeast growth phenotypes from pairwise deletions of non-essential yeast genes, determining affected areas in the hierarchy.

“This hierarchical approach was used to accurately predict yeast growth phenotypes from pairwise deletions of non-essential yeast genes, determining affected areas in the hierarchy.”

• cites Yu et al., 2016Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Programs have been developed to analyze and visualize cross-linking data as a network of amino acid nodes, where edges represent a detected cross-link, and organizing clusters of nodes spatially according to subunit is one approach applied in these networks.

“For example, programs have been developed to analyze and visualize cross-linking data as a network of amino acid nodes, where edges represent a detected cross-link (Combe et al., 2015; Courcelles et al., 2017; Grimm et al., 2015; Riffle et al., 2016); organizing clusters of nodes spatially according to subunit is one approach that has been applied in these networks (Figure 5C).”

• cites Combe et al., 2015 sourceNot checked yet

  Not checked in the free run.

• cites Courcelles et al., 2017 sourceNot checked yet

  Not checked in the free run.

• cites Grimm et al., 2015 sourceNot checked yet

  Not checked in the free run.

May combine two claims.

Genetic mutations may exert their effects at the level of the gene, the protein complex, or the broader pathway or organelle, and determining the relevant scale in the biological hierarchy can enable accurate prediction of phenotype.

“For example, genetic mutations in a genotype may exert their effects at the level of the gene, the protein complex, or the broader pathway or organelle; determining the relevant scale in the biological hierarchy can enable accurate prediction of phenotype, revealing the importance of multiple scales in modeling biological function.”

Probably not a substantial claim: may not pass S2 (checkable).

Hierarchical multiscale representations have the major advantages of generality across analytical methods and their scales and a comprehensive scope.

“On the other hand, the hierarchical multiscale representations have the major advantages of generality across analytical methods (and their scales) and a comprehensive scope (genome-wide, see Box 1).”

May combine two claims.

DeepGO used features from protein sequences and interactions to predict protein function; the features were used to build a hierarchical neural network classifier based on GO, so that predictions and features could be refined at each level of the biological hierarchy, resulting in a model optimized for function prediction.

“DeepGO used features from protein sequences and interactions to predict protein function; the features were used to build a hierarchical neural network classifier based on GO, so that predictions and features could be refined at each level of the biological hierarchy, resulting in a model optimized for function prediction (Kulmanov et al., 2018).”

• cites Kulmanov et al., 2018 sourceNot checked yet

  Not checked in the free run.

May combine two claims.

In the Yu et al. analysis, many genetic interactions were predicted within the oxidative phosphorylation subsystem; negative interactions occurred between the two subsystems of electron and protein transport, whereas positive interactions occurred solely within electron transport.

“For example, many genetic interactions were predicted within the oxidative phosphorylation subsystem; negative interactions (worse than expected growth) occurred between the two subsystems of electron and protein transport, whereas positive interactions (better than expected growth) occurred solely within electron transport.”

• cites Yu et al., 2016Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

One strategy for integrating hierarchical maps with spatial organization is to overlay the localization coordinates of each entity in 2D or 3D.

“One strategy for integrating hierarchical maps with spatial organization is to overlay the localization coordinates of each entity in 2D or 3D.”

May combine two claims.

Clustering analysis of networks is sensitive to network noise, and biochemical experimental challenges as well as cell-to-cell variation can contribute to this issue.

“Clustering analysis of networks is sensitive to network noise (Stacey et al., 2020), and biochemical experimental challenges as well as cell-to-cell variation can contribute to this issue.”

• cites Stacey et al., 2020 sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

Community detection algorithms such as Louvain or Newman-Girvan determine the approximate number of communities by optimizing modularity, or the density of links within communities versus links between communities.

“Community detection algorithms such as Louvain (Blondel et al., 2008) or Newman Girvan (Girvan and Newman, 2002) determine the approximate number of communities by optimizing modularity, or the density of links within communities versus links between communities.”

• cites Blondel et al., 2008 sourceNot checked yet

  Not checked in the free run.

• cites Girvan and Newman, 2002 sourceNot checked yet

  Not checked in the free run.

May combine two claims.

A complementary multiscale visualization strategy is to integrate 3D structural information with associated nodes in a hierarchy; software tools have been developed that display available 3D structural data for selected proteins and physical pairwise interactions in protein-protein interaction networks.

“A complementary multiscale visualization strategy is to integrate 3D structural information with associated nodes in a hierarchy; software tools have been developed that display available 3D structural data for selected proteins and physical pairwise interactions in protein-protein interaction networks (Morris et al., 2007; Mosca et al., 2013; Nepomnyachiy et al., 2015).”

• cites Morris et al., 2007 sourceNot checked yet

  Not checked in the free run.

• cites Mosca et al., 2013 sourceNot checked yet

  Not checked in the free run.

• cites Nepomnyachiy et al., 2015 sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

Another well-known problem in community detection is the tradeoff between quality control and completeness, since not all biological entities of interest may be detected or meet a quality threshold due to experimental challenges.

“Another well-known problem in community detection is the tradeoff between quality control and completeness, since not all biological entities of interest may be detected or meet a quality threshold due to experimental challenges.”

Probably not a substantial claim: may not pass S4 (based).

Various experimental data types and analytical methods generate networks of proximity between biological entities at different levels of the hierarchy and thus at different scales.

“Various experimental data types and analytical methods generate networks of proximity between biological entities at different levels of the hierarchy and thus at different scales.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Some approaches have used machine learning to integrate pairwise distance measurements at different scales.

“Some approaches have used machine learning to integrate pairwise distance measurements at different scales (Qin et al., 2020)”

• cites Qin et al., 2020Not checked yet

  Not checked in the free run.

Classified as probably not a substantial claim.

Many of the multiscale models described in the paper are presented as a static snapshot of biological structure.

“While many of the multiscale models that have been described herein are presented as a static snapshot of biological structure”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

Collectively, these techniques provide the opportunity to create unified hierarchical models across multiple relevant biological scales, from the very small (angstroms) to the very large (meters).

“Collectively, these techniques provide the opportunity to create unified hierarchical models across multiple relevant biological scales, from the very small (angstroms) to the very large (meters).”

Probably not a substantial claim: may not pass S2 (checkable).

Collectively, these approaches enable a unified hierarchical approach to biological data, with application from the molecular to the macroscopic.

“Collectively, these approaches enable a unified hierarchical approach to biological data, with application from the molecular to the macroscopic.”

Probably not a substantial claim: may not pass S3 (load-bearing).

When performing community detection at multiple scales to develop a hierarchy, there is the additional challenge of determining a suitable number of levels for the hierarchy, as this number is also typically not known beforehand.

“When performing community detection at multiple scales to develop a hierarchy, there is the additional challenge of determining a suitable number of levels for the hierarchy, as this number is also typically not known beforehand.”

May combine two claims.

Protein diversity, including sequence isoforms and post-translational modifications, controls interactions between proteins and individual protein function.

“in reality protein diversity (including sequence isoforms and post-translational modifications) controls interactions between proteins and individual protein function (Aebersold et al., 2018).”

• cites Aebersold et al., 2018 sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

In integrative structural biology, data from multiple methods are collected and translated to spatial restraints, which are used in scoring sampled structures.

“In integrative structural biology, data from multiple methods are collected and translated to spatial restraints, which are used in scoring sampled structures (Rout and Sali, 2019)”

• cites Rout and Sali, 2019Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

Some approaches have used machine learning to integrate single cell data with context-independent data to create cell-type specific interactomes.

“or to integrate single cell data with context-independent data to create cell-type specific interactomes (Mohammadi et al., 2019).”

• cites Mohammadi et al., 2019 sourceNot checked yet

  Not checked in the free run.