1 Computational Toxicology Promotes Regulatory Science
5
When using research to guide toxicology regulation, reproducible and rigorous
analysis are absolutely required. Bench experiments often have many variables which
are difficult to control, including variations in technician, machinery, laboratories,
reagent lots, reagent age, or other protocol subtleties. Advances in computer science
offer not only faster experiments, but also more reproducible ones. For example, a
computational analysis can be exactly repeated by an independent scientist provided
that the raw data is available, code is captured in a publicly available source such as
GitHub and the compute environment is dockerized. The ease of sharing experiments
not only allows computational toxicology to be more rigorous and reproducible, but
fosters collaboration between researchers as protocols are readily shared.
1.4 Methods in Computational Toxicology
Many computational techniques, including the ones originated from other fields
such as computational chemistry and pure computer science, have been developed
and applied in toxicology. To summarize, this book solicited chapters to review some
popular methods in computational toxicology that can be used to assess risk, evaluate
safety, and/or predict toxicology of a drug or other substance.
Chapter 2 introduced the modeling framework of computational toxicology,
defined its scope, listed the major tasks, reviewed the methods, and discussed the
challenges in computational toxicology.
Structural alerts and quantitative structure–activity relationship (QSAR) models
are two of the most popular methods for predicting toxicological activity of chemicals, especially for the simple toxicological endpoints [18, 19]. Chapter 3 reviewed
the applications of structural alerts and QSAR models in computational toxicology
and summarized some lessons learned from some successful models. It also discussed some challenges such as making negative predictions, moving to quantitative
predictions and weight of evidence approaches.
Emerging technologies such as next-generation sequencing enable fast generation of huge amounts of data. Computational analysis is challenging and crucial to
extract knowledge from such big data [20]. Machine learning algorithms have been
developed and applied in computational toxicology for prediction of unexpected,
toxic effects of chemicals. Moreover, computer science has enabled computational
prediction to scale to supermassive sizes. For example, the field of machine learning
has birthed matrix and tensor factorization. These two approaches have been used
to analyze >2.5 × 108 data points spanning 1300 compounds. It would be absolutely impossible to analyze such a dataset in a simple traditional program such as
Microsoft Excel! Chapter 4 reviewed the recent progresses in machine learning-based
computational methods and tools and further detailed matrix and tensor factorization
approaches.
One feature of modern science is diverse data for a specific scientific question
such as specific risk of chemicals to humans and the environment. Thus, integrating diverse data sources from toxicological research to extract more consistent and
5
When using research to guide toxicology regulation, reproducible and rigorous
analysis are absolutely required. Bench experiments often have many variables which
are difficult to control, including variations in technician, machinery, laboratories,
reagent lots, reagent age, or other protocol subtleties. Advances in computer science
offer not only faster experiments, but also more reproducible ones. For example, a
computational analysis can be exactly repeated by an independent scientist provided
that the raw data is available, code is captured in a publicly available source such as
GitHub and the compute environment is dockerized. The ease of sharing experiments
not only allows computational toxicology to be more rigorous and reproducible, but
fosters collaboration between researchers as protocols are readily shared.
1.4 Methods in Computational Toxicology
Many computational techniques, including the ones originated from other fields
such as computational chemistry and pure computer science, have been developed
and applied in toxicology. To summarize, this book solicited chapters to review some
popular methods in computational toxicology that can be used to assess risk, evaluate
safety, and/or predict toxicology of a drug or other substance.
Chapter 2 introduced the modeling framework of computational toxicology,
defined its scope, listed the major tasks, reviewed the methods, and discussed the
challenges in computational toxicology.
Structural alerts and quantitative structure–activity relationship (QSAR) models
are two of the most popular methods for predicting toxicological activity of chemicals, especially for the simple toxicological endpoints [18, 19]. Chapter 3 reviewed
the applications of structural alerts and QSAR models in computational toxicology
and summarized some lessons learned from some successful models. It also discussed some challenges such as making negative predictions, moving to quantitative
predictions and weight of evidence approaches.
Emerging technologies such as next-generation sequencing enable fast generation of huge amounts of data. Computational analysis is challenging and crucial to
extract knowledge from such big data [20]. Machine learning algorithms have been
developed and applied in computational toxicology for prediction of unexpected,
toxic effects of chemicals. Moreover, computer science has enabled computational
prediction to scale to supermassive sizes. For example, the field of machine learning
has birthed matrix and tensor factorization. These two approaches have been used
to analyze >2.5 × 108 data points spanning 1300 compounds. It would be absolutely impossible to analyze such a dataset in a simple traditional program such as
Microsoft Excel! Chapter 4 reviewed the recent progresses in machine learning-based
computational methods and tools and further detailed matrix and tensor factorization
approaches.
One feature of modern science is diverse data for a specific scientific question
such as specific risk of chemicals to humans and the environment. Thus, integrating diverse data sources from toxicological research to extract more consistent and