BioLizard provides the best of both worlds: biological knowledge + computational expertise
Presentation at AI4 the health value chain - 24th September 2020
Last week our Director of business development, Liesbeth Ceelen, presented BioLizard at the AI4Diag event, AI4 The health value chain, hosted by flanders.health. She gave a brief overview of the range of expertise we have at BioLizard, specifically our proven experience in machine learning and artificial intelligence.
Liesbeth emphasized that, “For us at BioLizard, it is crucial to bring together both biological data and AI.” Two of the major applications where we apply machine learning and AI tools are firstly for multi-omics data analysis/multimodality testing where we incorporate molecular data, clinical data as well as imaging data, and secondly for algorithm development.
During the presentation, Liesbeth provided some high-level examples where BioLizard has implemented machine learning and AI for specific client cases. One case focused on building predictive models for actual experimental data, while the other explored existing datasets using Natural Language Processing in order to generate a combination of models that would suggest optimally fitted sequences for clinical use.
Our approach for both cases was to first to establish accurate and robust predictive models through integration of different data sources and by implementing filtering normalization, data transformation, feature engineering methods, in order to assess where we could.
“For us at BioLizard, it is crucial to bring together both biological data and AI.”
Our workflow was more of an iterative process where we implemented various data transformations techniques and tested them using our panel of models to assess the predictive power for each dataset. Furthermore, we took into account and took steps to improve the completeness of the dataset, the highly volatile nature of the data, dataset size, as well as the fact that the data may be highly dimensional. All this together motivated the implementation and assessment of the various data transformation techniques to deal with this specific dataset.
All in all, our results provided our clients with robust models to be used for further studies and indicated the type of data needed for improved predictive capacity. We also highlighted important caveats to avoid in future experimental designs.
Concluding her remarks, Liesbeth stated that, “Machine learning based models are very useful to learn more about what scientific information is hidden in your data. However, machine learning is not magic and a proper data set size, quality of data and experimental design are crucial when developing predictive models.”
Finally, it is important to note that understanding the biology behind a research question is very important when applying machine learning based approaches, as it helps improve overall context of the problem at hand and can guide us into making more robust predictions.
If you would like to know more about BioLizard and our expertise, please contact us here.