Prediction of Allergenicity of Novel and Modified Proteins. TNO has developed a n in silico model to predict allergenicity of proteins based on Random Forest modelling. This tool can be applied in combination with extensive knowledge on allergen risk assessment and management in support of your (novel) food dossier as well as in early selection of product leads in NPD.
2. KEY CHALLENGES FOOD & NUTRITION
Allergenicity prediction tool
How to feed 9 B people by 2050? With foods grown in a ecologically
sustainable way?
Novel foods, new proteins and sources, such as insects, novel
vegetables etc
How to develop food proteins with improved functionality?
Processing and any other technologies, GMO
Lots of innovations, leading to new proteins or protein-containing
products in development
HOW ABOUT HEALTH RISKS?
3. HEALTH RISKS
Allergenicity prediction tool
Toxicological AllergenicNutritional Microbiological
Clear guidance and available tests
Limited guidance
and tests
4. ALLERGENICITY RISK
Cross-reactivity vs de novo sensitization
Allergenicity prediction tool
Cross reactivity
Is the novel (processed) protein able to elicit an
allergic reaction in a food allergic population (cross
reactivity, IgE binding)?
De novo sensitization
Is the novel (processed) protein able to induce a new
allergy (sensitization, IgE production)?
5. ALLERGENICITY RISK ASSESSMENT
Allergenicity prediction tool
Verhoeckx et al.,
Reg. Toxicol. Pharmacol. (79) 2016
No history
New TNO method
BIG CHALLENGE
for Novel/modified
Proteins
6. Machine learning as a tool for predicting the allergenicity of new food proteins:
Random Forest model developed to distinguish allergenic from “non-allergenic
proteins”
Over half a million proteins from protein databases. Proteins assigned as allergenic
when present in the allergen database COMPARE (n = 1680)
Protein characteristics from databases (SwissProt, UniProt) and mathematical
programs (e.g. ProtParam, PSIpred)
Model predicts with > 85% accuracy if a protein is an allergen or not.
BIG DATA SOLUTION
Westerhout et al., Reg. Toxicol. Pharmacol. 107 (2019)
8. ALL PROTEINS WERE CORRECTLY PREDICTED!
Name Species
Sequence
similarity with
known allergens
Predicted
allergen with
Random Forest
prediction model
Allergenic
Larval cuticle protein A2B Tenebrio molitor N Y Y
Larval cuticle protein A1A Tenebrio molitor N Y Y
Larval cuticle protein A3A Tenebrio molitor N Y Y
Alpha-amylase Tenebrio molitor Y Y Y
Tropomyosin-1, isoforms
9A/A/B
Drosophila
melanogaster
Y Y Y
Arginine kinase
Drosophila
melanogaster
Y Y Y
Cytochrome b Tenebrio molitor N N N
Elongation of very long chain
fatty acids protein
Tenebrio molitor N N N
TEST CASE: INSECT PROTEINS
Westerhout et al., Reg. Toxicol. Pharmacol. 107 (2019)
10. KEY APPLICATIONS
Screening novel or modified proteins for potential
allergenicity risks
Support in paragraph on allergenicity of novel food
proteins or products in EFSA/FDA safety dossiers
Selection for further investigation (e.g. IgE binding)
Risk-based candidate selection tool for novel or
modified proteins in early development programs
Allergenicity prediction tool
11. INTERESTED TO DISCUSS YOUR NEEDS?
PLEASE CONTACT:
Bas Kremer
Business developer Healthy Living
E: bas.kremer@tno.nl
P: +31 (0)88 866 6968