Data Preparation

• The process of collecting, cleaning, and organizing large datasets of variants along with their corresponding pathogenicity annotations

• Critical for ensuring the quality and reliability of the model’s predictions

graph TD; 1(Data collection) 1 -- Raw data - missense variants + respective protein sequences--> 2 2(Data annotation) 2 -- Data annotated with pathogenicity/benign status--> 3 3(Cleaned data) 3 -- Removing duplicated and erroneous data --> 4 4(Split data) --> 5(Training data) 4(Split data) --> 6(Validation data) 4(Split data) --> 7(Test data) A(Ensure the variants are accurately identified <br> and annotated with their respective protein sequences) --> B(Gathering pathogenicity annotations <br> from reliable sources) B --> C(Ensures the data is clean,<br> consistent, and ready to be used for modeling) C --> D(Model training/validation and testing) classDef highlight fill:#99ccff; classDef white fill:#ffffff; class 1,2,3,4,5,6,7 highlight;

Note

Dataset Splitting :mega:

• Training data:

  • Train the AlphaMissense model

  • Bening: missense variants frequently observed in human and primate populations

  • Pathogenic: missense variants absent from human and primate populations

• Validation data:

  • Tune model parameters

  • Held-out data - Pathogenic missense variants in ClinVar (ClinVar ) & Bening variants from population-databases

• Test data:

  • Evaluate the model’s performance on unseen data

  • Held-out data - Pathogenic missense variants in ClinVar and other selected studies & Bening variants from population-databases

Attention

Validation data vs test data

• Validation data:

  • Used during model training to guide many model-building decisions

• Test data:

  • Completely unseen dataset used at the very end to get an unbiased evaluation of the fully trained model