Automatic workflow for in vitro high-throughput screening data FAIRification, preprocessing and scoring.
Creators
- 1. Ideaconsult Ltd., University of Plovdiv, Faculty of Chemistry, Department of Analytical Chemistry and Computer Chemistry, Bulgaria
- 2. Karolinska Insitutet, Institute of Environmental Medicine, Sweden
- 3. Misvik biology, Division of Toxicology, Finland
- 4. Ideaconsult Ltd.
- 5. Ideaconsult, Bulgaria
Description
The development of new chemical substances and advanced materials, including nanomaterials (NMs), pose complex challenges to ensure safety for humans and environment. Regulatory agencies are interested in adopting safety data generated under the umbrella term “New Approach Methodologies” (NAM) encompassing technologies for high-throughput, efficient integration of experimental data, QSAR and read across reports during innovation, from an idea to a chemical substance/NM product launch. Data management based on FAIR (Findability, Accessibility, Interoperability, and Reuse) guiding principles supports consistent curation and reuse of the accumulated data by the nanosafety, cheminformatics and bioinformatics communities. In vitro high-throughput screening (HTS) hazard data is used for efficient clustering, ranking, prioritization of NMs and read across. A previously developed in-vitro toxicity scoring and ranking concept, the Tox5Score [1] is applied in two stages: (i) normalization of the HTS metrics, separately in the range [0-1], for each time point and endpoint; (ii) combination of the normalized metric values to obtain final Tox5 endpoint scores. However, the usage of Excel based data preprocessing with the application of the software ToxPi (the US-EPA Toxicological Prioritization Index) requires time consuming manual processing, which is hard to scale up for larger NM datasets and occasionally prone to errors.
Here we present an automated workflow for data FAIRification, preprocessing and calculation of the Tox5Score from raw HTS data. A new Python module for collection and annotation of raw data, consequent normalization and calculation of dose-response metrics was developed. The module invokes ToxPi-R library and strictly follows the original Tox5 approach. The module can be used independently or as a part of developed by us Orange [2] workflow with custom widgets for fine tuning of the data processing. The Orange (open source system for visual programming and machine learning [2]) workflow includes separate widgets for data normalization, dose-response calculation, Tox5 in-vitro toxicity scoring, ranking for specific cell, visualization of ToxScore for endpoint- and time-point-specific toxicity, ranks, and combined toxicity scores for each material. The widget’s table output can be exported in convenient file formats (e.g. CSV).
In addition, the new Python module and Orange workflow extends the eNanoMapper FAIRification workflow [3] by facilitating FAIRification of HTS data. The resulting FAIR data includes both raw and interpreted data (scores) in machine readable format and can be distributed as data archive and/or be integrated into the eNanoMapper database and Nanosafety Data Interface [4].
- Nymark, P; Hongisto, V et al. Toxicology Letters, 314, 2019, https://doi.org/10.1016/j.toxlet.2019.09.002
- Demsar, J et al, Journal of Machine Learning Research, 2013, 2349−2353.
- Kochev, N et al. Nanomaterials, 10, 2020, https://doi.org/10.3390/nano10101908
- Jeliazkova, N et al. Nat. Nanotechnol. 16, 2021, 644–654 https://doi.org/10.1038/s41565-021-00911-6
Files
poster_GA_HARMLESS_jan_2024.pdf
Files
(1.4 MB)
Name | Size | Download all |
---|---|---|
md5:954d0c80fd6e58e68e63b69bbfb1ec3f
|
1.4 MB | Preview Download |