19 OpenTox Principles and Best Practices for Trusted Reproducible …
391
Reproducibility of a testing strategy would therefore encompass the reproducibility of the experimental data as well as the algorithmic or in silico model. Relying
on a documented test protocol to conduct the experiment and keeping track of any
deviations from the original protocol is a key. A knowledge management system
where test protocols are tracked and versioned is a good practice. OECD publishes
harmonized templates that describe the data formats for reporting information that
is used for the risk assessment of chemicals. Its applications involve, not only “studies done on chemicals to determine their properties or effects on human health and
the environment, but also for storing data on use and exposure.” The harmonized
templates are aimed at developers of database systems, as they prescribe the formats by which information can be entered into and maintained in a database. The
test study summary information can be exchanged between governments, regulators, and industry electronically using these templates. The templates can also be
used to report summary test results for different products. REACH registrants use
the OHTs for submitting their results to the European Chemicals Agency (ECHA)
using IUCLID. On the other hand, the theoretical model behind the ITS can vary
from simple decision trees on the order by which risk assessors should perform their
experiments [7] to more complicated machine learning algorithms [8] or Bayesian
networks [9].
ITS skin sensitization example: The biology of skin sensitization is described
by an adverse outcome pathway (AOP) involving a sequence of key events representing potential biological activity responses to chemical exposure of the skin (see
Fig. 19.3). Validated assays against the key events may be used in combination with
in silico methods for an ITS purpose. Within the OECD framework, a guideline exists
for Integrated Approaches to Testing and Assessment (IATAs) which consist of an
ITS where different strategies or algorithms may be used, as described by a “defined
approach.” We have recently developed an ITS application for skin sensitization (see
Fig. 19.4) [10] reproducing the IATA and defined approach developed at Procter
and Gamble and involving Bayesian networks [9]. The use of Bayesian networks
allows tolerating missing information and conveys the probabilistic hypothesis of
skin sensitization based on accumulative evidence from data. In building the ITS,
open-source tools were favoured by Douglas Connect (now renamed Edelweiss Connect (EwC)). A public open-access reference application is made available for the
scientific community to evaluate at https://its.douglasconnect.com.
19.5 Best Practices in Data Management: FAIR Principles,
Data Completeness, Interoperability and Ontology
Comprehensive documentation of the scientific procedures, experimental as well
as computational, is of uttermost importance to achieve reproducibility. Besides
guidelines for running the experiments including good laboratory practice and cell
culture practice guides (which are out of scope of this current paper), detailed test
391
Reproducibility of a testing strategy would therefore encompass the reproducibility of the experimental data as well as the algorithmic or in silico model. Relying
on a documented test protocol to conduct the experiment and keeping track of any
deviations from the original protocol is a key. A knowledge management system
where test protocols are tracked and versioned is a good practice. OECD publishes
harmonized templates that describe the data formats for reporting information that
is used for the risk assessment of chemicals. Its applications involve, not only “studies done on chemicals to determine their properties or effects on human health and
the environment, but also for storing data on use and exposure.” The harmonized
templates are aimed at developers of database systems, as they prescribe the formats by which information can be entered into and maintained in a database. The
test study summary information can be exchanged between governments, regulators, and industry electronically using these templates. The templates can also be
used to report summary test results for different products. REACH registrants use
the OHTs for submitting their results to the European Chemicals Agency (ECHA)
using IUCLID. On the other hand, the theoretical model behind the ITS can vary
from simple decision trees on the order by which risk assessors should perform their
experiments [7] to more complicated machine learning algorithms [8] or Bayesian
networks [9].
ITS skin sensitization example: The biology of skin sensitization is described
by an adverse outcome pathway (AOP) involving a sequence of key events representing potential biological activity responses to chemical exposure of the skin (see
Fig. 19.3). Validated assays against the key events may be used in combination with
in silico methods for an ITS purpose. Within the OECD framework, a guideline exists
for Integrated Approaches to Testing and Assessment (IATAs) which consist of an
ITS where different strategies or algorithms may be used, as described by a “defined
approach.” We have recently developed an ITS application for skin sensitization (see
Fig. 19.4) [10] reproducing the IATA and defined approach developed at Procter
and Gamble and involving Bayesian networks [9]. The use of Bayesian networks
allows tolerating missing information and conveys the probabilistic hypothesis of
skin sensitization based on accumulative evidence from data. In building the ITS,
open-source tools were favoured by Douglas Connect (now renamed Edelweiss Connect (EwC)). A public open-access reference application is made available for the
scientific community to evaluate at https://its.douglasconnect.com.
19.5 Best Practices in Data Management: FAIR Principles,
Data Completeness, Interoperability and Ontology
Comprehensive documentation of the scientific procedures, experimental as well
as computational, is of uttermost importance to achieve reproducibility. Besides
guidelines for running the experiments including good laboratory practice and cell
culture practice guides (which are out of scope of this current paper), detailed test