In-silico protocols for reproducibility of machine learning approaches

This poster shows our in-silico protocols approach to improve reproducibility of Machine Learning (ML). It builds on top of SMART Protocols for wet-lab experiments to adapt them and extend them to cover Machine Learning experiments. Lab-protocols often accompany wet-lab experiments as text-based documents describing the sequence of tasks and operations executed, including, e.g., references to equipment, reagents troubleshooting and tips. SMART Protocols provides a semantic layer for lab-protocols so, for instance, reagents are linked to (semantic) chemical databases such as ChEMBL. Similarly to wet-lab experiments, ML experiments are also composed of inputs, steps and outputs and would benefit from semantically enriched in-silico protocols. We aim at providing a semantic layer, namely in-silico protocol, for ML experiments supporting FAIRness for ML, transparency and reproducible outcomes. More information can be found at GitHub Repository. Funding This work is partially funded by the German Research Foundation (DFG) under the grant No. 460234259 corresponding to NFDI4DataScience consortium.