Projects of AIMS

Click on a project to find out more details:

Inter-Project Collaborations

AIDIA-AIDEX

In connection with the AIDEX sub-project (Dr. Teiser), video files will be available that show the movement of granular particles, which often move quickly and often obscure each other. In the course of the AIDIA project, a recording configuration is to be developed in cooperation with AIDEX that enables a 3D reconstruction that is as highly accurate and seamless as possible. This will be presented and realized within the framework of AIDEX, and corresponding AI methods for evaluation will be developed and made available at AIDIA. AI-based (deep learning) solutions for the depth-from-defocus task are to be adapted to the AIDEX particle experiments. A corresponding methodological building block can be useful for improving further solutions for camera-based 3D reconstruction or can be reused for their construction. All developed building blocks are tested in close cooperation with the working group of Prof. Dr. Wurm and AIDEX and adapted to corresponding applications. In the cooperation with AIDEX, close coordination with Dr. Teiser is planned from the outset.
Finally, it will also be investigated by way of example – although this is not of central importance for AIDIA during the development phase – whether the methods developed can be used to investigate the dust dynamics from the experiments of the AIPEX project (Prof. Dr. Schwarz).

AIDIA-AMCRIS

In the AMCRIS project (Prof. Dr. Grimm, Dr. Schulz, Dr. Wehland), dynamically changing, multicellular structures are displayed in image stacks from FLUMIAS. Here, the dynamics are observed over a longer period of time in order to recognize special events such as spheroid formation and to obtain a detailed 3D model for the process of metastasis formation. For the recognition of 3D structures, a volumetric variant of CNNs appears to be a first candidate for the approach in the project, which is generally very complex and requires a large amount of training data even by the standards of deep learning. Depending on the data and initial more precise analyses, it could be useful for recognizing the formation of spheroids and other important structures to segment the movement in the images based on the optical flow instead of the image content itself. This would then amount to a so-called motion segmentation task, which in favorable cases can be easily analyzed using relatively simple AI methods.

AIDIA-AIDEX

In connection with the AIDEX sub-project (Dr. Teiser), video files will be available that show the movement of granular particles, which often move quickly and often obscure each other. In the course of the AIDIA project, a recording configuration is to be developed in cooperation with AIDEX that enables a 3D reconstruction that is as highly accurate and seamless as possible. This will be presented and realized within the framework of AIDEX, and corresponding AI methods for evaluation will be developed and made available at AIDIA. AI-based (deep learning) solutions for the depth-from-defocus task are to be adapted to the AIDEX particle experiments. A corresponding methodological building block can be useful for improving further solutions for camera-based 3D reconstruction or can be reused for their construction. All developed building blocks are tested in close cooperation with the working group of Prof. Dr. Wurm and AIDEX and adapted to corresponding applications. In the cooperation with AIDEX, close coordination with Dr. Teiser is planned from the outset.
Finally, it will also be investigated by way of example – although this is not of central importance for AIDIA during the development phase – whether the methods developed can be used to investigate the dust dynamics from the experiments of the AIPEX project (Prof. Dr. Schwarz).

AIPEX-AIDEX-AIGE

The universal applicability of the developed optical systems is to be evaluated and optimized in collaboration with the AIPEX and AIGE projects. Specifically, it is planned to integrate the developed image acquisition methods into various experimental setups and to evaluate their applicability to other systems. In the case of the collaboration with the AIPEX project, systems for 3D image acquisition can be tested in experiments with complex plasmas. In the case of the collaboration with the AIGE project, these are experiments on the dynamics of excited granular gases. It is also planned to test the developed optical systems in combination with existing experimental setups (AIPEX: complex plasmas, AIGE: excited granular gases) as part of joint experiments.

AIDIA-ASIM0V

In the experiments from ASIMOV (Dr. Böhmer), the first task in connection with the tracking of plastids is to register the microscopic geometries three-dimensionally in high quality. For this purpose, a three-dimensional version of AI-based optical flow models is to be developed that performs the registration (and which is reused for tasks arising in AMCRIS). Depending on the specific data situation, a volumetric CNN could serve as the basis for this, but simpler, locally operating AI methods also appear to be a possible basis for registration if the displacements to be determined between the image stacks are not too large.

DAIMLER-ASIM0V

The project objective of DAIMLER is the generation of artificial data sets to provide evaluation AIs (e.g. a CNN) with sufficiently large training data. A GAN is being developed that can generate statistical outliers in measurement data. The data from ASIM0V is particularly suitable for achieving the objectives, as the data set is small and therefore cannot actually be used for training CNNs. A maximum of 8 root tips can be observed during the experiments in the microfluidic channel. The resulting data volume is in the range of less than 100 images. In addition, there are statistical outliers in the calcium concentration, which must also be recorded during training. However, this is not possible due to the limited data available. Artificial data must therefore supplement the training set.

ASIM0V-AIPEX

The data prioritization algorithms that are being developed for ASIMOV in the DAIMLER project are to be transferred to data from AIPEX. Due to the geometric similarity of the measurement data (starch grains, complex plasma as a point cloud), there are interdisciplinary synergies here. The experience gained from ASIM0V is to be incorporated into the adaptation of the reduction and prioritization algorithms.

DAIMLER-AIPEX

If the GAN is successfully implemented in cooperation with ASIM0V, the software will be qualified on PK4 data in order to prepare an evaluation AI for the follow-up experiment COMPACT. This will be done in close cooperation with Prof. Thoma and Prof. Schwarz. The training data will be provided by AIPEX in the form of images of complex plasmas. Specifically, artificial experimental data for PK4 will be generated using the GAN method. This includes the provision of experimental data to the HS Mittweida, which mirrors 3 sigma scattering of the experimental data back into the AIPEX project using the GAN method. This data can be used to take into account extreme cases of experimental evaluation that do not occur in regular experimental procedures. Using this data, the AI algorithms can then be better trained and consequently achieve better performance.

DAIMLER-AMCRIS

The aim of AMCRIS and the joint partners DAIMLER and AIDIA is the development and application of AI image analysis techniques for the evaluation of image data recorded under real µg with the high-resolution fluorescence microscope FLUMIAS on the ISS. This includes the identification and reliable quantification of molecular structures and organelles. In addition, AI methods will be made available to recognize and classify tubular and spheroid structures that are as resolution-independent as possible. This also includes a study on the 3D reconstruction of spheroidal structures based on Z-stacks.

IAI-XPRESS-AMCRIS

In a pilot study, RNAseq data obtained under r-µg and s-µg will be analyzed using deep learning algorithms and their analytical linkage with general (ENCODE, RoadMap, GEO etc.) and own epigenetic data obtained under long-term µg. In this pilot project, the already obtained RNAseq data of the prostate cancer cell lines DU145 (moderate metastatic potential) and PC-3 (high metastatic potential) will be analyzed integratively under µg with prioritization via epigenetic information of the cell lines. The data preparation, normalization and filtering is carried out under the supervision of Dr. Herbert Schulz, the subsequent evaluation is carried out by the AI partners XTRAS and IAI-XPRESS. To solve the classification task, the LVQ models developed and used in IAI-XPRESS detect relevant genes for class differentiation or correlations between gene expressions that contribute to class differentiation (methods of relevance and classification correlation learning for LVQ). The evaluation/interpretation of the resulting relevance profiles or correlations of the genes then allows the medical-biological scientists in the AMCRIS project to draw conclusions about the sensitivity of the genes with regard to gravitational conditions. If necessary, medical expert knowledge can be integrated into the LVQ procedures in order to improve interpretability. This must be realized in close technical cooperation with the AMCRIS team.ASIM0V-AIPEXThe data prioritization algorithms that are being developed for ASIMOV in the DAIMLER project are to be transferred to data from AIPEX. Due to the geometric similarity of the measurement data (starch grains, complex plasma as a point cloud), there are interdisciplinary synergies here. The experience gained from ASIM0V is to be incorporated into the adaptation of the reduction and prioritization algorithms.

XTRAS-ASIM0V

In the classification of changes under microgravity, explainable CNNs are developed by XTRAS to analyze the image data sets generated and provided by ASIM0V from the FLUMIAS microscope. Here, the analysis to discriminate signal transduction waves is about spatial and temporal changes of calcium, ROS or pH values at the single cell level using MISAT and GECO experiments at different g-forces. The CNN will recognize patterns within the different signals and use them for classification according to the two resolutions and make them extractable. Furthermore, XTRAS is working on the development of modular neural networks for the analysis and integration of different omics expression datasets to analyze the importance of biomarkers on regulation under microgravity. For this purpose, omics data sets from ASIM0V will be generated or have already been generated from other campaigns to create regulatory networks. Subsequently, the identified biomarkers will be transferred from XTRAS back to ASIM0V to generate knockout mutants for a validation approach.

DAIMLER – AIDIA – XTRAS – IAI-XPRESS

This joint project aims to achieve close methodological integration between the AI experts. AIDIA and DAIMLER are working closely together methodically in the area of feature extraction and data reduction. Synergies are already known from joint project experience in GeoFlow. DAIMLER and XTRAS both work with CNNs. The methodological separation results from the specific choice of networks. While XTRAS specializes in “explainable deep networks” (XAI), DAIMLER works primarily with interpretable networks. The cooperation results from the application of biological data that is to be analyzed with both network types. IAI-XPRESS works with smart interpretable networks, which are a methodological complement to XTRAS. In cooperation with IAI-XPRESS and XTRAS, various AI and XAI models from the field of neural networks and smart interpretable AI algorithms are developed and compared for the respective expression data sets from the groups of Prof. Grimm and Dr. Böhmer. The data for gene expression analysis are provided or alternatively analyzed in the projects AMCRIS (Prof. Grimm) and ASIMOV (Dr. Böhmer). This makes it possible to compare the performance with the methods provided in IAI-XPRESS. Human cell cultures and plant roots will be analyzed using different AI methods. The two biological research groups are supported by DAIMLER in data generation and data reduction and the AIDIA group helps with the processing of the image data sets and the improvement of NN models.

XTRAS-AMCRIS

A modular NN with an explainable component such as LRP is to be developed for AMCRIS in order to analyze the omics expression data sets obtained under r-µg and s-µg according to the importance of individual biomarkers. Proteomics as well as RNA-Seq and Methyl-Seq datasets will be generated or have already been generated by AMCRIS. These datasets will be used to identify biomarkers to distinguish microgravity or tumor-specific metabolic pathways. In addition, the generated image data sets of AMCRIS from the FLUMIAS microscope can be analyzed for “Clever Hans predictors” to classify between spheroid formation or metastasis under microgravity. Here, the extracted important features can be used for discrimination to identify specific features of the metastasis model.

AIGE-AMCRIS

In the AIGE project, the extraction of structures from FLUMIAS microscope data will also be investigated, especially for granular and general physical samples; for this purpose, AMCRIS will investigate how different samples can be mapped in the instrument to optimize the settings and assess the image and data quality. In this cooperation, strategies are to be found to be able to use FLUMIAS in an interdisciplinary way and to evaluate how AI methods can be used to support data evaluation.

AIGE-DAIMLER-AIDIA

Prioritization methods and data reduction algorithms from the DAIMLER-AIPEX and DAIMLER-ASIM0V collaborations are to be applied to AIGE data in the final phase of the project. The effort required to adapt the developed algorithms in order to analyze the experimental data of granulate packs in a follow-up phase of AIMS will be examined. The 3D tracking algorithms of the AIDIA project will also be tested for applicability. The aim of the cooperation is a feasibility study on the portability and applicability of the developed image processing AI methods.

ASIM0V-AMRICS-XTRAS

Both life sciences groups ASIM0V and AMCRIS have already conducted different campaigns of drop tower, parabolic flight and sounding rocket experiments to generate different temporally resolved transcriptomics and proteomics datasets. These omics datasets with a small number of replicates normally exhibit different batch effects due to environmental changes and thus influence the prediction of microgravity biomarkers. XTRAS will use modular NN with LRP to analyze the omics datasets from ASIM0V and AMCRIS and create integrative multi-omics networks. This will be used by all groups in close collaboration to identify regulatory biomarkers and metabolic pathways under microgravity that are either present in both plants and humans or are species-specific.