Multimodal/Multidimensional Data Annotation and Model Construction

DataCicada will extend its platform Tympana for multimodal and multidimensional data (such as multiomics analysis). Updating Tympana to handle multimodal and multidimensional datastreams will allow researchers to scale their time by requiring less effort on low level data annotation tasks and the subsequent Machine Learning (ML) model construction from that data, reducing the cycle time from data collection to deployed model. Tympana, a 2022 Phase I SBIR awardee, is currently geared towards single dimensional recognition tasks in Imaging, Sensor or Omics type domains. However, complex workflows in the biological sciences often include diverse multimodal/multidimensional sets of data that can be combined in interesting and novel ways. Extending Tympana to allow for combination of datasets that include metadata, imaging, omic information and time series allows for greater construction of model types for custom predictive tasks. For example, the workflow for spatial transcriptomics requires the ability to slice the data through multiple dimensions quickly and easily to determine spatially resolved gene expression. While highly dimensional in itself, the data can also be accompanied by experimental metadata that can also serve as input to a ML model. By updating the platform to handle multiple co-occurring data streams, Tympana will be able to extend to new cutting-edge research domains. Users of Tympana do not need to be experts in Machine Learning (ML) to benefit from the power of automated recognition. Tympana can recognize the patterns of interest within a sample. This pattern of interest is provided by the subject matter expert in a quick and intuitive interface. Models are iteratively trained as the user works with the data, building trust in the system. Once ready, models are easily exportable in containers for integration into analytic pipelines increasing the scalability of domain expertĺs time. By putting the researcher in the driverĺs seat, Tympana allows for rapid development of custom ML models without requiring integration with a ML team. This reduces the barrier in scaling up analytic workflows from weeks to hours. By quickly building custom models and deploying them for unique cases, researchers can focus on larger questions. DataCicada brings expertise in fundamental sciences, Machine Learning and Data Science, having extensive experience analyzing diverse data streams and the common challenges across them. DataCicada will complete the work covered in the scope of this proposal in the United States.