A spectral graph theoretic approach to quantification and calibration of collective morphological differences in cell images

Abstract

Motivation: High-throughput image-based assay technologies can rapidly produce a large number of cell images for drug screening, but data analysis is still a major bottleneck that limits their utility. Quantifying a wide variety of morphological differences observed in cell images under different drug influences is still a challenging task because the result can be highly sensitive to sampling and noise.

Results: We propose a graph-based approach to cell image analysis. We define graph transition energy to quantify morphological differences between image sets. A spectral graph theoretic regularization is applied to transform the feature space based on training examples of extremely different images to calibrate the quantification. Calibration is essential for a practical quantification method because we need to measure the confidence of the quantification. We applied our method to quantify the degree of partial fragmentation of mitochondria in collections of fluorescent cell images. We show that with transformation, the quantification can be more accurate and sensitive than that without transformation. We also show that our method outperforms competing methods, including neighbourhood component analysis and the multi-variate drug profiling method by Loo et al. We illustrate its utility with a study of Annonaceous acetogenins, a family of compounds with drug potential. Our result reveals that squamocin induces more fragmented mitochondria than muricin A.

Availability: Mitochondrial cell images, their corresponding feature sets (SSLF and WSLF) and the source code of our proposed method are available at http://aiia.iis.sinica.edu.tw/.

Supplementary information: Supplementary data are available at Bioinformatics online.

Fig. 1.

Representative microscopic fluorescent images of single cells with different levels of fragmentation of mitochondria. Intact mitochondria forms filamentous networks (top row), completely fragmented mitochondria have round shape (bottom row) and partially fragmented mitochondria (middle row).

Fig. 2.

Given two sets of cell images with extremely different morphology patterns as training examples; our method transforms the feature space so that regularized graph transition energy between the two sets is minimized. Then we can quantify the morphological difference of cells by computing their graph transition energy with the training examples in the transformed feature space.

Fig. 3.

The results of Mito-Q image processing. (a) The original image; (b) the image processed by the morphological filter; (c) the image after power-law transformation; (d) the locations of mitochondrial identified by Mito-Q.

Fig. 4.

(a) Influence of the k nearest neighbors. (b) The performance of semi-supervised learning before and after learning κ in the HeLa cell image data set.

Fig. 5.

Visualizing shape similarity of 10 distinct subcellular organelles with a MDS plot derived from pair-wise graph transition energy.

Fig. 6.

Influence of the k-NN.

Fig. 7.

Distributions of the similarity of profiles as proposed by Loo et al. (2007) (a) and distributions of energy estimated by the graph-based method with (c) and without (b) feature space transformation for the 1st (pure MI, 0% MC), 6th (mixture of MI and MC with equal proportions, 50% MC) and 11th (pure MC, 100% MC) treatment sets.

Fig. 8.

KL divergence of the distributions of the quantification of the difference between control (pure MI) and the treatment sets with increasing proportion of MC.

Fig. 9.

Comparing the effect of mitochondrial fragmentation by squamocin and muricin A in CHO cells.

Fig. 10.

Representative images of the cells after treated by DMSO (a), muricin A (b) and squamocin (c) for a certain number of days.

Fig. 11.

Morphological similarity of mitochondria in different image groups by MDS according to the pairwise graph transition energy values.