LICTOR (λ-LIght-Chain TOxicity predictoR) is a machine learning approach able to classify lambda (λ) immunoglobulin light chains (LCs), as either toxic or non-toxic, depending on their likelihood to form toxic species inducing light chain amyloidosis disease. LICTOR uses as input LCs protein sequences. From an evaluation conducted on 1,075 LC sequences, LICTOR obtained an area under the receiver operating characteristic (AUC) of 0.87, with Specificity (Sp) and Sensitivity (Se) values of 0.82 and 0.76, respectively. Sensitivity and specificity were computed maximizing the Youden index (J) as a function of the confidence level, i.e., the probability that a sequence belongs to the predicted phenotype. J is frequently used as a further analysis of the ROC (Receiver Operating Characteristic) curve. J measures the effectiveness of a diagnostic test and allows the estimation of the optimal threshold value (c) for the test, which is the value where Se (c) + Sp (c) -1 is maximized. Maximizing J index we found that the optimal threshold value for LICTOR, in identifying toxic light chain sequences, is 0.46.

Try it!