Limits...
A novel phenotypic dissimilarity method for image-based high-throughput screens.

Zhang X, Boutros M - BMC Bioinformatics (2013)

Bottom Line: Here we propose a new analysis method, PhenoDissim, which computes the phenotypic dissimilarity between cell populations via Support Vector Machine classification and cross validation.PhenoDissim is a novel analysis method for image-based high-throughput screen, relying on two parameters which can be automatically optimized without a priori knowledge.PhenoDissim is freely available as an R package.

View Article: PubMed Central - HTML - PubMed

Affiliation: German Cancer Research Center (DKFZ), Div, Signaling and Functional Genomics and Department of Cell and Molecular Biology, Medical Faculty Mannheim, Im Neuenheimer Feld 580, D-69120 Heidelberg, Germany. xianzhang@gmail.com.

ABSTRACT

Background: Discovering functional relationships of genes through cell-based phenotyping has become an important approach in functional genomics. High-throughput imaging offers the ability to quantitatively assess complex phenotypes after perturbation by RNA interference (RNAi). Such image-based high-throughput RNAi screening studies have facilitated the discovery of novel components of gene networks and their interactions. Images generated by automated microscopy are typically analyzed by extracting quantitative features of individual cells, resulting in large multidimensional data sets. Robust and sensitive methods to interpret these data sets and to derive biologically relevant information in a high-throughput and unbiased manner remain to be developed.

Results: Here we propose a new analysis method, PhenoDissim, which computes the phenotypic dissimilarity between cell populations via Support Vector Machine classification and cross validation. Applying this method to a kinome RNAi screening data set, we demonstrate that the proposed method shows a good replicate reproducibility, separation of controls and clustering quality, and we are able to identify siRNA phenotypes and discover potential functional links between genes.

Conclusions: PhenoDissim is a novel analysis method for image-based high-throughput screen, relying on two parameters which can be automatically optimized without a priori knowledge. PhenoDissim is freely available as an R package.

Show MeSH

Related in: MedlinePlus

A network of identified phenotypes and their phenotypic dissimilarity. 31 siRNAs are identified as phenotype hits and calculated for pair-wise phenotypic dissimilarity. Each siRNA is represented by a node. siRNA pairs with phenotypic dissimilarity smaller than 0.82 are connected with an edge, with only connected nodes shown. Cell images for representative phenotypes are shown and labeled with the corresponding siRNAs.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4225524&req=5

Figure 3: A network of identified phenotypes and their phenotypic dissimilarity. 31 siRNAs are identified as phenotype hits and calculated for pair-wise phenotypic dissimilarity. Each siRNA is represented by a node. siRNA pairs with phenotypic dissimilarity smaller than 0.82 are connected with an edge, with only connected nodes shown. Cell images for representative phenotypes are shown and labeled with the corresponding siRNAs.

Mentions: In total, 31 siRNA perturbations (averaging two replicates of each gene) showing high phenotypic dissimilarity to siRLUC control (>0.85) indicate morphological phenotypes. With the pair-wise phenotypic dissimilarity for the 31 siRNA samples, we generated a network of phenotypes with nodes representing each phenotype and edges for phenotype dissimilarity between nodes as in Figure 3 (only phenotypic dissimilarity smaller than 0.82 and connected nodes are shown), as well as representative cell images. From network connectivity and visual inspection, we found three major groups of phenotypes. Genes highlighted in green are essential genes, and cause viability defect when knocked down. Within this group are genes PLK1 and COPB2, but also other genes such as PKM2 and PMVK. Genes highlighted in blue cause cell shape defect when depleted by siRNA. Cells are often elongated with thin stretches, suggesting defect in cell structure maintenance. Genes highlighted in orange cause strong actin staining and also affect cell shape. Genes in gray show intermediate phenotypes between the major groups. For example, siRAC1 treated cells show both a slight viability defect and an elongated shape. Further experiments to explain the underlying basis of these phenotypes are needed, however, in some cases previous functional characterizations support the observed phenotypes and their mechanism. For example, MRC2 was previously shown to be responsible for the turn-over of collagen [27] and higher levels of collagen was associated with elongated cell shapes [28]. TESK2 was shown to be involved in actin cytoskeletal organization [29]. It should also be noted that the same morphology phenotype can be caused by unrelated mechanisms, nevertheless, grouping similar phenotypes may help identify and understand functionally related genes and their interactions.


A novel phenotypic dissimilarity method for image-based high-throughput screens.

Zhang X, Boutros M - BMC Bioinformatics (2013)

A network of identified phenotypes and their phenotypic dissimilarity. 31 siRNAs are identified as phenotype hits and calculated for pair-wise phenotypic dissimilarity. Each siRNA is represented by a node. siRNA pairs with phenotypic dissimilarity smaller than 0.82 are connected with an edge, with only connected nodes shown. Cell images for representative phenotypes are shown and labeled with the corresponding siRNAs.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4225524&req=5

Figure 3: A network of identified phenotypes and their phenotypic dissimilarity. 31 siRNAs are identified as phenotype hits and calculated for pair-wise phenotypic dissimilarity. Each siRNA is represented by a node. siRNA pairs with phenotypic dissimilarity smaller than 0.82 are connected with an edge, with only connected nodes shown. Cell images for representative phenotypes are shown and labeled with the corresponding siRNAs.
Mentions: In total, 31 siRNA perturbations (averaging two replicates of each gene) showing high phenotypic dissimilarity to siRLUC control (>0.85) indicate morphological phenotypes. With the pair-wise phenotypic dissimilarity for the 31 siRNA samples, we generated a network of phenotypes with nodes representing each phenotype and edges for phenotype dissimilarity between nodes as in Figure 3 (only phenotypic dissimilarity smaller than 0.82 and connected nodes are shown), as well as representative cell images. From network connectivity and visual inspection, we found three major groups of phenotypes. Genes highlighted in green are essential genes, and cause viability defect when knocked down. Within this group are genes PLK1 and COPB2, but also other genes such as PKM2 and PMVK. Genes highlighted in blue cause cell shape defect when depleted by siRNA. Cells are often elongated with thin stretches, suggesting defect in cell structure maintenance. Genes highlighted in orange cause strong actin staining and also affect cell shape. Genes in gray show intermediate phenotypes between the major groups. For example, siRAC1 treated cells show both a slight viability defect and an elongated shape. Further experiments to explain the underlying basis of these phenotypes are needed, however, in some cases previous functional characterizations support the observed phenotypes and their mechanism. For example, MRC2 was previously shown to be responsible for the turn-over of collagen [27] and higher levels of collagen was associated with elongated cell shapes [28]. TESK2 was shown to be involved in actin cytoskeletal organization [29]. It should also be noted that the same morphology phenotype can be caused by unrelated mechanisms, nevertheless, grouping similar phenotypes may help identify and understand functionally related genes and their interactions.

Bottom Line: Here we propose a new analysis method, PhenoDissim, which computes the phenotypic dissimilarity between cell populations via Support Vector Machine classification and cross validation.PhenoDissim is a novel analysis method for image-based high-throughput screen, relying on two parameters which can be automatically optimized without a priori knowledge.PhenoDissim is freely available as an R package.

View Article: PubMed Central - HTML - PubMed

Affiliation: German Cancer Research Center (DKFZ), Div, Signaling and Functional Genomics and Department of Cell and Molecular Biology, Medical Faculty Mannheim, Im Neuenheimer Feld 580, D-69120 Heidelberg, Germany. xianzhang@gmail.com.

ABSTRACT

Background: Discovering functional relationships of genes through cell-based phenotyping has become an important approach in functional genomics. High-throughput imaging offers the ability to quantitatively assess complex phenotypes after perturbation by RNA interference (RNAi). Such image-based high-throughput RNAi screening studies have facilitated the discovery of novel components of gene networks and their interactions. Images generated by automated microscopy are typically analyzed by extracting quantitative features of individual cells, resulting in large multidimensional data sets. Robust and sensitive methods to interpret these data sets and to derive biologically relevant information in a high-throughput and unbiased manner remain to be developed.

Results: Here we propose a new analysis method, PhenoDissim, which computes the phenotypic dissimilarity between cell populations via Support Vector Machine classification and cross validation. Applying this method to a kinome RNAi screening data set, we demonstrate that the proposed method shows a good replicate reproducibility, separation of controls and clustering quality, and we are able to identify siRNA phenotypes and discover potential functional links between genes.

Conclusions: PhenoDissim is a novel analysis method for image-based high-throughput screen, relying on two parameters which can be automatically optimized without a priori knowledge. PhenoDissim is freely available as an R package.

Show MeSH
Related in: MedlinePlus