236
P. Volarath et al.
12.2 Computational Approaches Currently Used in Food
Ingredient Safety Assessments
In this section, we describe basic principles and applications of some computational
methods that are currently utilized for safety assessment of food ingredients, such as
in silico models and databases. While validation “Standard Operating Procedures”
and similar aspects of “Good Practices” are crucial for scientific accountability in
safety assessments, these aspects are not covered in depth here. It is expected that
practitioners will adhere to quality control and assurance procedures in performing
computational modeling [6].
12.2.1 QSAR Modeling
QSAR modeling represents a subclass of in silico models that predicts potential toxicity of chemicals using molecular descriptors. Molecular descriptors are computercalculated chemical and physicochemical features that describe the chemical structures and properties, respectively. Examples of descriptors commonly used in QSAR
modeling are summarized in Table 12.1 [7]. Different descriptors have advantages
and disadvantages, and the selection of descriptors should be based on the type of
models that need to be built [8, 9]. Once appropriate descriptors are selected, an
algorithm (such as linear regression or multiple linear regression) can be applied to
generate a QSAR model. In general, chemicals whose activities fit the same QSAR
model are assumed to have similar biological/biochemical functions and are, therefore, expected to behave through the same mechanism and exert similar toxicity.
In OFAS, QSAR models have been used to support premarket safety assessments
and provide guidance to industry and other stakeholders during prenotification consultations (PNCs) for future food contact substance notifications (FCNs) [10]. For
example, an FCN must include all data and other information that form the basis
of the determination that the FCS is safe under the intended conditions of use. Data
must include primary toxicological [4] and chemical [2] information. Typically, with
respect to the chemical data this includes information about residual starting material, catalysts, adjuvants, production aids, byproducts and breakdown products of the
FCS. FDA encourages petitioners and notifiers to contact the agency before making
a submission to discuss various issues related to a submission. Prior to submitting an
FCN, industry can submit a PNC to discuss regulatory and scientific aspects of their
intended submission, including the data needed to support the FCN. In response to a
PNC, OFAS review scientists routinely use QSAR models to identify potential safety
questions related to genetic toxicity and carcinogenicity for substances with no or
limited safety data. In some cases, certain toxicological endpoints, such as developmental and reproductive toxicity, may be evaluated using QSAR models. Thus, upon
review of a PNC, the FDA can recommend the types of toxicity data needed to come
to a safety conclusion based on the results of the QSAR analyses. Review scientists’
P. Volarath et al.
12.2 Computational Approaches Currently Used in Food
Ingredient Safety Assessments
In this section, we describe basic principles and applications of some computational
methods that are currently utilized for safety assessment of food ingredients, such as
in silico models and databases. While validation “Standard Operating Procedures”
and similar aspects of “Good Practices” are crucial for scientific accountability in
safety assessments, these aspects are not covered in depth here. It is expected that
practitioners will adhere to quality control and assurance procedures in performing
computational modeling [6].
12.2.1 QSAR Modeling
QSAR modeling represents a subclass of in silico models that predicts potential toxicity of chemicals using molecular descriptors. Molecular descriptors are computercalculated chemical and physicochemical features that describe the chemical structures and properties, respectively. Examples of descriptors commonly used in QSAR
modeling are summarized in Table 12.1 [7]. Different descriptors have advantages
and disadvantages, and the selection of descriptors should be based on the type of
models that need to be built [8, 9]. Once appropriate descriptors are selected, an
algorithm (such as linear regression or multiple linear regression) can be applied to
generate a QSAR model. In general, chemicals whose activities fit the same QSAR
model are assumed to have similar biological/biochemical functions and are, therefore, expected to behave through the same mechanism and exert similar toxicity.
In OFAS, QSAR models have been used to support premarket safety assessments
and provide guidance to industry and other stakeholders during prenotification consultations (PNCs) for future food contact substance notifications (FCNs) [10]. For
example, an FCN must include all data and other information that form the basis
of the determination that the FCS is safe under the intended conditions of use. Data
must include primary toxicological [4] and chemical [2] information. Typically, with
respect to the chemical data this includes information about residual starting material, catalysts, adjuvants, production aids, byproducts and breakdown products of the
FCS. FDA encourages petitioners and notifiers to contact the agency before making
a submission to discuss various issues related to a submission. Prior to submitting an
FCN, industry can submit a PNC to discuss regulatory and scientific aspects of their
intended submission, including the data needed to support the FCN. In response to a
PNC, OFAS review scientists routinely use QSAR models to identify potential safety
questions related to genetic toxicity and carcinogenicity for substances with no or
limited safety data. In some cases, certain toxicological endpoints, such as developmental and reproductive toxicity, may be evaluated using QSAR models. Thus, upon
review of a PNC, the FDA can recommend the types of toxicity data needed to come
to a safety conclusion based on the results of the QSAR analyses. Review scientists’