Machine learning predicts immunoglobulin light chain toxicity through somatic mutations

In light chain (AL) amyloidosis, pathogenic monoclonal light chains (LCs) deposit as amyloid fibrils in target organs. Molecular determinants of LC pathogenicity are currently unknown. Here, we present LICTOR, a method to predict LC toxicity based on the distribution of somatic mutations acquired during clonal selection. LICTOR achieves specificity and sensitivity of 0.82 and 0.76, respectively, with an AUC of 0.87, making it a valuable tool for early AL diagnosis.