AlphaMissense versus laboratory-based pathogenicity prediction of 13 novel missense CPA1 variants from pancreatitis cases

We have read with great interest the study by Wang et al 1 in which the authors evaluated the utility of the AlphaMissense prediction programme2 (https://alphamissense.hegelab.org) in the classification of missense CPA1 variants with respect to pathogenicity in chronic pancreatitis. While the AI-driven prediction performed relatively well, the authors highlighted potential shortcomings that can limit its value in clinical practice. Defining the pathogenic potential of CPA1 variants detected in pancreatitis cases can be challenging because the mechanistic basis of disease risk is unrelated to loss of CPA1 function and seems to be determined by mutation-induced misfolding and the ensuing endoplasmic reticulum (ER) stress.3–5 Recently, we used transiently transfected HEK 293T cells to measure the secretion efficiency and induction of BiP mRNA expression, a marker of ER stress, for 50 missense CPA1 variants from pancreatitis cases and healthy controls.6 We found that the best predictor of pathogenicity was loss of secretion (<10% of wild type) irrespective of BiP levels. This data set can serve as a reference for the assignment of clinical significance of novel CPA1 variants. In the present study, we set out to examine what fraction of novel CPA1 variants detected in real-world genetic testing can be classified as pathogenic and whether AlphaMissense can replace …