Skip to main content
. 2024 May 15;10(20):eadl0161. doi: 10.1126/sciadv.adl0161

Fig. 4. Validate the predictive power of RACER-m with external datasets.

Fig. 4.

(A) Prediction results of RACER-m on the HLA-A*02:01–restricted systems from 10x Genomics dataset collected from five healthy donors. A total of 1741 unique pairs of TCR-peptide sequences were tested, and the prediction results of z score were grouped by the immunological profile of the test TCR-pMHC pairs and depicted as box plots. (B) Comparison of classification performance between RACER-m and NetTCR-2.0 (11) on a curated list of public TCR-pMHC repertoires (12, 42) composed of both strong binders and mismatched weak binder. Because of the restriction of NetTCR-2.0 on the peptide length (9-mer), there are no data from NetTCR-2.0 for the two 10-mer peptides (KLVALGINAV and ELAGIGILTV). (C) The classification performance of RACER-m on another set of TCR-pMHC test TCR-pMHC pairs (34). AUROC, area under the ROC curve.