Phil Harrison presented a poster at Swedish Symposium on Deep Learning, 2021

Fluorescence imaging is a valuable tool for biological analysis but is time-consuming and toxic to the cells. Using deep learning to virtually stain bright-field images is an active field of research that can alleviate these problems. Phil Harrison, a PhD student in the HASTE group, presented a poster at the Swedish Symposium on Deep Learning (SSDL) 2021 based on the HASTE team’s winning solution for the Adipocyte Cell Imaging Challenge. The poster presented our approach and results.

HASTE Team wins the Adipocyte Cell Imaging Challenge 2020 organised by AstraZeneca and AI Sweden

AI Sweden and AstraZeneca organised the Adipocyte Cell Imaging Challenge, a two-week-long hackathon to help AstraZeneca accelerate the drug development process. The task was to use machine learning in solving the problem of labelling cell images without requiring toxic preprocessing of cell cultures by predicting the content of the fluorescence images from the corresponding bright-field images.

Eight teams consisting of Swedish and international representatives from both academia and the private sector participated in the challenge. Our winning team consisted of Ankit Gupta, Håkan Wieslander, Phil Harrison, Ebba Bergman from HASTE Team and Erik Hallström from Wählby lab. The winner was selected by a jury consisting of representatives from AstraZenecaVinnova and AI Sweden.

To the left is the maximum projection of 7 brightfield images. In the middle, the corresponding superimposed fluorescence image of cell nuclei (blue), lipid droplets (green), and cytoplasm (red). To the right is the result of our proposed solution.

The solution used the Learning Under Privileged Information (LUPI) paradigm to solve the problem. LUPI enables the inclusion of additional (privileged) information when training machine learning models, data that is not available when making predictions. In this case, the segmentation masks of the nuclei were used as the privileged information during the training of machine learning models. Our solution will help AstraZeneca to speed up the drug discovery process and bring drugs to market quicker.

More information on challenge can be found here.

Press release about the solution can be found here, here, and here.

HASTE project meeting over zoom.

Unfortunately, we didn’t get a chance to enjoy the good food and beautiful environment of the Noor Castle, but we had a productive project meeting anyway. The meeting started with report reading and Salman Toor presenting his docent lecture “Distributed Computing e-Infrastructures (DCI): Challenges and Opportunities for Application Experts and Service Providers”. This was followed by intense brainstorming on the continued research in the project, and a discussion on the almost complete half-time report, and planning for the coming Tuesday seminars, which have become an important part of the project communication now that most of us work from home.

HASTE meeting over Zoom

Hongru Zhai joins the HASTE team to work on the developing hierarchical representation of the microscopy image data

 

Hongru is a master’s student from the department of statistics, and his main interests in statistics include multivariate statistical methods and Bayesian statistics.

Hongru’s MSc thesis will focus on developing the better hierarchical representations of the microscopy image data from cellular experiments with the help of statistical methods, focusing on improving readability and informational efficiency of the representation.

Tianru Zhang joins HASTE team as new PhD Student to work on management of large data streams

We welcome Tianru Zhang as new PhD Student in the Hellander lab at the Department of Information Technology, Uppsala University.

Tianru obtained his Bachelor in Probability and Statistics in Mathematics at the University of Science and Technology of China in 2017. Then, he completed his Master in Statistics for Smart Data at ENSAI (The National School of Statistics and Analysis of Information of France) in 2018. Before moving to Uppsala, he was employed as Assistant Researcher at the Fujitsu R&D center Co., Ltd. where he worked on developing DeepTensor (a deep learning method using tensor decomposition) and analyzing data of personal online loans.

Andrea Behanova joins the HASTE team to work on the enhancement of image quality by registration of short exposure miniTEM images

Andrea is doing a traineeship at the Department of Information Technology – Division of Visual Information and Interaction, Uppsala University while coursing the last semester from the master’s in Medical Physics at the University of Eastern Finland.

Andrea’s internship project focus on developing an approach of registering and aggregating short exposure miniTEM images (superresolution reconstruction). The project objective is to achieve better quality and higher resolution image compare to a long exposure one.

Ebba Bergman joins HASTE team as PhD student

We welcome Ebba Bergman as new PhD Student in the Spjuth lab at Department of Pharmaceutical Biosciences, Uppsala University.

Ebba obtained her Master of Science in Engineering, with a focus on bioinformatics, from Uppsala University in 2017. Before starting her PhD Ebba worked as a full-stack systems developer for 2 years. 

About the PhD project within HASTE:  
Ebba is currently working on Conformal Prediction (CP) in combination with Convolutional Neural Networks. Next, she will focus on applying Learning Under Privileged Information (LUPI) on transmission electron microscopy data provided by Vironova. In general, Ebba will focus on combining machine learning methods with CP and LUPI using data provided by our HASTE-project partners.

Presentation at COPA 2019

Ola Spjuth, Co-PI in the HASTE project, presented two accepted HASTE-papers at the [8th Symposium on Conformal and Probabilistic Prediction with Applications](http://clrc.rhul.ac.uk/copa2019) in Varna, Bulgaria on 9-11 Sept 2019. The two papers below are now published in [Proceedings of Machine Learning Research (PMLR) volume 105](https://proceedings.mlr.press/v105/).

Paper 1: Split Knowledge Transfer in Learning Under Privileged Information Framework

Gauraha, N., Söderdahl, F. and Spjuth, O.
Split Knowledge Transfer in Learning Under Privileged Information Framework. 
Proceedings of Machine Learning Research (PMLR). 105, 43-52. (2019).
ABSTRACT
Learning Under Privileged Information (LUPI) enables the inclusion of additional (privileged) information when training machine learning models, data that is not available when making predictions. The methodology has been successfully applied to a diverse set of problems from various fields. SVM+ was the first realization of the LUPI paradigm which showed fast convergence but did not scale well. To address the scalability issue, knowledge transfer approaches were proposed to estimate privileged information from standard features in order to construct improved decision rules. Most available knowledge transfer methods use regression techniques and the same data for approximating the privileged features as for learning the transfer function. Inspired by the cross-validation approach, we propose to partition the training data into K folds and use each fold for learning a transfer function and the remaining folds for approximations of privileged features—we refer to this as split knowledge transfer. We evaluate the method using four different experimental setups comprising one synthetic and three real datasets. The results indicate that our approach leads to improved accuracy as compared to LUPI with standard knowledge transfer.

Paper 2: Combining Prediction Intervals on Multi-Source Non-Disclosed Regression Datasets

Spjuth O., Brännström R.C., Carlsson L. and Gauraha, N.
Combining Prediction Intervals on Multi-Source Non-Disclosed Regression Datasets.
Proceedings of Machine Learning Research (PMLR). 105, 53-65. (2019).
ABSTRACT
Conformal Prediction is a framework that produces prediction intervals based on the output from a machine learning algorithm. In this paper, we explore the case when training data is made up of multiple parts available in different sources that cannot be pooled. We here consider the regression case and propose a method where a conformal predictor is trained on each data source independently, and where the prediction intervals are then combined into a single interval. We call the approach Non-Disclosed Conformal Prediction (NDCP), and we evaluate it on a regression dataset from the UCI machine learning repository using support vector regression as the underlying machine learning algorithm, with a varying number of data sources and sizes. The results show that the proposed method produces conservatively valid prediction intervals, and while we cannot retain the same efficiency as when all data is used, efficiency is improved through the proposed approach as compared to predicting using a single arbitrarily chosen source.