Fig 1.
The 10 tips for establishing AI or ML model validity.
Curation tips encompass considerations for designing both training and validation sets. Processing tips pertain to the methodology for feature selection, metric calibration, and batch correction/normalization of data. Validation tips focus on the methods used to assess the quality of learning exhibited by models.
Fig 2.
Curation tips covering tips 1 to 4.
Fig 3.
How batch effects (A) and cross-sample normalization (B) can confound ML.
Fig 4.
Validation tips covering null model benchmarks (A) and validating the trained model extensively on as many distinct (heterogeneous) dataset as possible (B).