Home   |   ACC    |  HTS Browser  |  Contact Us                     ACC Wiki

News | Description | Download | Video tutorial | Publications


ACC workshop in Lausanne, 29. April 2010

We would like to invite you to the second Advanced Cell Classifier (ACC) and High-Content Data Chain (HCDC) workshop, in Lausanne EPFL 29. April 2010; (location will be announced).

Give a deeper insight into advanced-level high-content screening and data analysis and practical help to install and use the software being developed in the ETH Light Microscopy Centre/RISC.


Global session
10:00-10.45: Advanced Cell Classifier, machine learning techniques, and statistical analysis of HC screens by Peter Horvath
10:45-11.30: High-Content Data Chain introduction by Karol Kozak
11.30-12.00: Pitfalls and screening in practice by Andreas Vonderheit

12.00-13.30: Lunch break

Session 1. (Data analysis track)
13.30-14.15: HCDC install and usage
14:15-15.00: ACC install and case study

Session 2. (Wet-work/general track)
13.30-13.55: Image based screening activities at EPFL by Gerardo Turcatti
13.55-14.20: Image based screening activities at ETH Zurich by Gabor Csucs
14.20-15.00: To be announced

Please write an email to  peter.horvath@lmc.biol.ethz.ch

Gerardo Turcatti (EPFL)
Peter Horvath (ETH)

Now the third genome-wide

I am happy to share that in the last 3 months we finished with the primary analysis of the second whole genome-wide screen and the third one is acquired and will happen to be analyzed in the next few weeks. Several other very interesting smaller siRNA and drug screens had been analyzed and we reach the 1 billion classified cells with Advanced Cell Classifier.


Semi-supervised learning module

Semi-supervised learning (SSL) is a method combining the opportunities of both supervised and unsupervised learning, using unlabeled data and some supervision information to better estimate classes than unsupervised learning but requiring less field expertise than in case of supervised learning. Our primary goal is to develop a module where the user defines the positive, negative and if other known controls are available than those, and running SSL clustering we try to homogenize classes and learn them. After the first trials we conclude that using the proposed method we were able to identify ~95% of the hits with less than 10 mouse click instead of several hours teaching of supervised classifiers.


First genome-wide screen with ACC

We finished the primary analysis of a human genome-wide (~22.000 genes, using 4 oligos) screen. The result shows the reliability of ACC, the achieved Z' factor is 0.755, which is considered as excellent. Technical details: more than 2 million fluorescent images; ~80 million cells were classified; 92% classification accuracy measured by cross validation. 

Collaboration and Masters thesis

Research Institutes and Masters students are welcomed to collaborate with us to further develop or customize Advanced Cell Classifier.

For Masters students the following topics are suggested but original ideas are welcomed:

  • Semi-supervised learning methods in high-content screening
    The aim of this project would be to develop and implement a new classification method and compare its accuracy to supervised methods on our large screen databases
    Background: programming skills and interest in bioinformatics (ETH students preferred, externals with own scholarship)
  • Statistical analysis and hit detection of cell-based screens
    The primary goal of this project is to better understand and detect interesting (unusual) events in cell-based high-content screens. The candidate should design a solid statistical approach for normalization and hit detection and implement it.
    Background: strong statistical and mathematical, basic programming knowledge required with interest in bioinformatics (ETH students preferred, externals with own scholarship)


For research institutes:

  • Joint projects in HCA, machine learning or higher-level statistical analysis
  • Customization of ACC bringing new features


If you are interested contact us.


Advanced Cell Classifier is a data analyzer program for High Content Screening experiment to more accurately identify different phenotypes. The biggest aim is to reduce user interaction but still preserve accuracy. The program is developed in MatLab, therefore it is platform independent.

System requirements:

  • MatLab 7.5 (R2007)
  • Image processing toolbox
  • Neural Networks toolbox
  • Matlab report generator

System properties:

  • 96 and 384 plate format
  • html report generation
  • several classification methods
  • quick image prediction
  • versatile image visualization

Available classification methods:

  • LibSVM
  • Neural Network
  • Logistic.Weka
  • MultilayerPerceptron.Weka
  • RBFNetwork.Weka
  • SimpleLogistic.Weka
  • SMO.Weka
  • RandomCommittee.Weka
  • RandomForest.Weka
  • BayesNet.Weka
  • NaiveBayes.Weka
  • AdaBoostM1.Weka
  • Bagging.Weka
  • Dagging.Weka
  • END.Weka
  • EnsembleSelection.Weka
  • LogitBoost.Weka
  • BFTree.Weka
  • FT.Weka
  • J48.Weka
  • RandomTree.Weka
  • REPTree.Weka


T. Wild, P. Horvath, E. Wyler, B. Widmann, L. Badertscher, I. Zemp, K. Kozak, G. Csusc, E. Lund, and U. Kutay. A protein inventory of human ribosome biogenesis reveals an essential function of Exportin 5 in 60S subunit export. 2010 PLoS Biology

P. Horvath, T. Wild, U. Kutay, G. Csucs. Machine Learning Improves the Precision and Robustness of High-Content Screens: Using Nonlinear Multiparametric Methods to Analyze Screening Results. 2011. J Biomol Screening


The system is implemented in MatLab with GUI.  

Documentation [AdvancedCellClassifier.pdf] (381 kb)
Advanced Cell Classifier 1.1 [ACCv1_1.zip] (5 203 kb)
Sample images [samples.zip] (2 926 kb)

If you have further questions contact us, we are happy to help you to install and insert into your current platform.

Video tutorial

Last update: 14.09.2011