Dataset Information

Evaluation of methods for generative modeling of cell and nuclear shape.

ABSTRACT: Motivation:Cell shape provides both a geometry for, and a reflection of, cell function. Numerous methods for describing and modeling cell shape have been described, but previous evaluation of these methods in terms of the accuracy of generative models has been limited. Results:Here we compare traditional methods and deep autoencoders to build generative models for cell shapes in terms of the accuracy with which shapes can be reconstructed from models. We evaluated the methods on different collections of 2D and 3D cell images, and found that none of the methods gave accurate reconstructions using low dimensional encodings. As expected, much higher accuracies were observed using high dimensional encodings, with outline-based methods significantly outperforming image-based autoencoders. The latter tended to encode all cells as having smooth shapes, even for high dimensions. For complex 3D cell shapes, we developed a significant improvement of a method based on the spherical harmonic transform that performs significantly better than other methods. We obtained similar results for the joint modeling of cell and nuclear shape. Finally, we evaluated the modeling of shape dynamics by interpolation in the shape space. We found that our modified method provided lower deformation energies along linear interpolation paths than other methods. This allows practical shape evolution in high dimensional shape spaces. We conclude that our improved spherical harmonic based methods are preferable for cell and nuclear shape modeling, providing better representations, higher computational efficiency, and requiring fewer training images than deep learning methods. Availability and Implementation:All software and data is available for reviewers at http://murphylab.cbd.cmu.edu/software/2018_CellShapeModels. This link is currently hidden (and not tracked) and will be made public upon manuscript acceptance by linking to it from http://murphylab.cbd.cmu.edu/software. Supplementary information:Supplementary data are available at Bioinformatics online.

SUBMITTER: Ruan X

PROVIDER: S-EPMC6612826 | biostudies-other | 2018 Dec

REPOSITORIES: biostudies-other

ACCESS DATA

Publications

Evaluation of methods for generative modeling of cell and nuclear shape.

Ruan Xiongtao X Murphy Robert F RF

Bioinformatics (Oxford, England) 20190701 14

<h4>Motivation</h4>Cell shape provides both geometry for, and a reflection of, cell function. Numerous methods for describing and modeling cell shape have been described, but previous evaluation of these methods in terms of the accuracy of generative models has been limited.<h4>Results</h4>Here we compare traditional methods and deep autoencoders to build generative models for cell shapes in terms of the accuracy with which shapes can be reconstructed from models. We evaluated the methods on dif ...[more]

PMID: 30535313

Similar Datasets

Project description:Accurate predictions of the DNA binding specificities of transcription factors (TFs) are necessary for understanding gene regulatory mechanisms. Traditionally, predictive models are built based on nucleotide sequence features. Here, we employed three- dimensional DNA shape information obtained on a high-throughput basis to integrate intuitive DNA structural features into the modeling of TF binding specificities using support vector regression. We performed quantitative predictions of DNA binding specificities, using the DREAM5 dataset for 65 mouse TFs and genomic-context protein binding microarray data for three human basic helix-loop-helix TFs. DNA shape-augmented models compared favorably with sequence-based models for these predictions. Although both k-mer and DNA shape features encoded the interdependencies between nucleotide positions of the binding site, using DNA shape features reduced the dimensionality of the feature space compared to k-mer use. Finally, analyzing the weights of DNA shape-augmented models uncovered TF family- specific structural readout mechanisms that were not obvious from the nucleotide sequence. Three genomic-context protein binding microarray (gcPBM) experiments of human transcription factors were performed. Briefly, the gcPBMs involved binding his-tagged transcription factors c-Myc, Max, and Mad1(Mxd1) to double-stranded 180K Agilent microarrays in order to determine their binding specificity for putative DNA binding sites in native genomic context. Briefly, we represent three categories of 36-bp sequences: 1) bound probes, 2) unbound probes (or negative controls), and 3) test probes. Bound probes corresponded to genomic regions bound in vivo by c-Myc, Max, or Mad2 (ChIP-seq P < 10^(-10) in HeLaS3 or K562 celld (ENCODE)) that contain at least two consecutive 8-mers with universal PBM E-score > 0.4 (Munteanu and Gordan, LNCS 2013). All putative binding sites occur at the same position within the probes on the array. M-bM-^@M-^\UnboundM-bM-^@M-^] probes corresponded to genomic regions with ChIP-seq P < 10^(-10) and a maximum 8-mer E-score < 0.2. We also designed test probes that contain, within constant flanking regions, all nnCACGTGnn 10-mers and 18 nnnCACGTGnnn 12-mers (where n = A, C, G, or T). Each DNA sequence represented on the array is present in 6 replicate spots. We report the gcPBM signal intensity for each spot. The PBM protocol is described in Berger et al., Nature Biotechnology 2006 (PMID 16998473).

Dataset Information

Evaluation of methods for generative modeling of cell and nuclear shape.

Publications

Evaluation of methods for generative modeling of cell and nuclear shape.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets