Towards an automatic analysis of CHO-K1 suspension growth in microfluidic single-cell cultivation

Abstract

Motivation Innovative microfluidic systems carry the promise to greatly facilitate spatio-temporal analysis of single cells under well-defined environmental conditions, allowing novel insights into population heterogeneity and opening new opportunities for fundamental and applied biotechnology. Microfluidics experiments, however, are accompanied by vast amounts of data, such as time series of microscopic images, for which manual evaluation is infeasible due to the sheer number of samples. While classical image processing technologies do not lead to satisfactory results in this domain, modern deep learning technologies such as convolutional networks can be sufficiently versatile for diverse tasks, including automatic cell counting as well as the extraction of critical parameters, such as growth rate. However, for successful training, current supervised deep learning requires label information, such as the number or positions of cells for each image in a series; obtaining these annotations is very costly in this setting. Results We propose a novel machine learning architecture together with a specialized training procedure, which allows us to infuse a deep neural network with human-powered abstraction on the level of data, leading to a high-performing regression model that requires only a very small amount of labeled data. Specifically, we train a generative model simultaneously on natural and synthetic data, so that it learns a shared representation, from which a target variable, such as the cell count, can be reliably estimated. Availability The project is cross-platform, open-source and free (MIT licensed) software. We make the source code available at https://github.com/dstallmann/cell_cultivation_analysis; the data set is available at https://pub.uni-bielefeld.de/record/2945513

Novel transfer learning schemes based on Siamese networks and synthetic data

“…Since the learning methodology is semi-supervised, our formerly introduced Twin-VAE [22] will be used as a basis to propose a novel transfer learning method to mitigate the aforementioned complications. Here, synthetic data are used as an auxiliary training set Syn which is also used for transfer learning.…”

“…This is done by drawing cells as ellipsoids, varying some attributes like their brightness, size and blurriness of edges. For further detail, see the original Paper [22]. Reconstructions of real appearances from synthetic data, while interesting to suggesting inherent stability, are not of importance for a high accuracy on the task.…”

“…-We improved performance and lowered computational complexity (outperforming the original work [22]) -We build an efficient transfer pipeline and showed on two microscopy datasets empirically that it outperformed a variety of methods, including state-of-the-art image processing. -By performing extensive ablations studies we gained insight into which parts of the architecture contains representations which are beneficial for the transfer learning ability of the network.…”

“…We particularly focus on mitigating human labor for annotations. Our previously introduced deep twin auto-encoder architecture Twin-VAE [22] is trained on data stemming from one imaging modality and thereafter transferred to the similar yet different domain of the other microscopy technology. This training procedure greatly reduces the need for natural, labeled data, by using synthetic, auxiliary training data, for which the ground truth is known and which is easy to obtain in this setting, since the Twin-VAE does not require the images to be rendered realistically in every regard, such as morphological details.…”

“…2. In addition, there exists no Deep Learning Models which easily and efficiently solve the task, as shown in [22] by comparing to EfficientNet [23], and Watershed methods [17] and shown here later by comparing to BigTransfer [11] and our previous work. The emergence of more data in such specialized domains like this makes it important to provide an easy-to-use system which has a high performance and enables automation of processes involving this kind of data.…”