The results of the AnDi challenge 2020 are in the article: Muñoz-Gil et al. Objective comparison of methods to decode anomalous diffusion, Nature Communications 12:6253 (2021).
Objective and tasks
The competition was aimed to assess the performance of methods in quantifying anomalous diffusion, with respect to three different tasks:
- T1 – anomalous exponent inference
- T2 – model classification
- T3 – trajectory segmentation
each for 1D, 2D, and 3D trajectories. The performance was assessed with common metrics on simulated datasets with trajectory length and signal-to-noise level reproducing realistic experimental conditions. The submitted methods were also compared on the blind analysis of experimental trajectories.
Participants
team/software | method | task1 | task2 | task3 | platform | open access | contact |
---|---|---|---|---|---|---|---|
Anomalous Unicorns HYDRAS | Ensemble of CNN and RNN | 1D | 1D | - | Python | GitHub | Borja Requena ICFO-The Institute of Photonic Sciences Castelldefels (Barcelona), Spain |
BIT | Bayesian inference | 1D+2D+3D | 1D+2D+3D | 1D+2D+3D | Matlab | GitHub | Michael A. Lomholt PhyLife, Department of Physics, Chemistry and Pharmacy, University of Southern Denmark Odense M, Denmark |
DecBayComp Gratin | Graph neural network | 1D | 1D | - | Python | GitHub | Jean-Baptiste Masson Institut Pasteur, Decision and Bayesian Computation lab Paris, France |
DeepSPT | ResNet + XGBoost | 1D | 1D | Python | GitHub | Taegeun Song Center for AI and Natural Sciences, Korea Institute for Advanced Study Seoul, Korea |
|
eduN RANDI | RNN + Dense NN | 1D+2D+3D | 1D+2D+3D | 1D+2D+3D | Python | GitHub | Stefano Bo Max Planck Institute for the Physics of Complex Systems (MPI-PKS) Dresden, Germany |
Erasmus MC FEST | bi-LSTM + Dense NN | 1D+2D+3D | 1D+2D+3D | Python | GitHub | Hélène Kabbech Erasmus MC, Department of Cell Biology Rotterdam, The Netherlands |
|
FCI | CNN | 1D+2D | 1D+2D | 1D+2D | Python | GitHub | Tom Bland The Francis Crick Institute London, UK |
HNU Just LSTM it | LSTM | 1D+2D+3D | Python | GitHub | Zihan Huang School of Physics and Electronics, Hunan University Changsha, China |
||
NOA | CNN + bi-LSTM | 1D | Python | GitHub | Jose Alberto Conejero Instituto Universitario de Matemática Pura y Aplicada, Universitat Politècnica de València Valencia, Spain |
||
QuBI AnDi-ELM | ELM | 1D | 1D | Matlab | GitHub | Carlo Manzo Facultat de Ciències i Tecnologia, UVIC-UCC Vic, Spain |
|
TSA | Scaling analysis and feature engineering | 1D+2D+3D | 1D | Python | GitHub | Erez Aghion Max Planck Institute for the Physics of Complex Systems (MPI-PKS) Dresden, Germany |
|
UCL CONDOR | Feature engineering + NN | 1D+2D+3D | 1D+2D+3D | Matlab | GitHub | Giorgio Volpe Department of Chemistry, University College London London, UK |
|
UPV-MAT | CNN + bi-LSTM | 1D+2D+3D | 1D+2D+3D | Python | GitHub | Òscar Garibo i Orts Instituto Universitario de Matemática Pura y Aplicada, Universitat Politècnica de València Valencia, Spain |
|
Wust ML A | 1D: RISE + forest classier 2D: MrSEQL + logistic reg. | 1D+2D+3D | Python | GitHub | Janusz Szwabiński Faculty of Pure and Applied Mathematics, Wrocław University of Science and Technology Wrocław, Poland |
||
Wust ML B | Gradient boosting regression + classifier | 1D+2D | 1D+2D+3D | Python | T1: GitHub T2: GitHub | Hanna Loch-Olszewska & Patrycja Kowalek Faculty of Pure and Applied Mathematics, Wrocław University of Science and Technology Wrocław, Poland |
Datasets
For the final phase of the competitions, methods were evaluated over the test dataset 2020. Datasets were simulated using the andi-datasets package.
Organizers
- Gorka Muñoz-Gil & Maciej Lewenstein, Quantum Optics Theory – ICFO
- Carlo Manzo, the QuBI lab – FCT, UVic-UCC
- Giovanni Volpe, Soft Matter Lab – University of Gothenburg
- Miguel A. Garcia-March, UPV
- Ralf Metzler, Theoretical Physics – UniPotsdam