Machine learning integrations for development of a T-cell-tolerance derived signature to improve the clinical outcomes and precision treatment of hepatocellular carcinoma.

Hepatocellular carcinoma (HCC) is characterized by high rates of recurrence and metastasis and poor prognosis. A recently discovered concept of T cell tolerance (TCT) has become an entirely new target of cancer immunotherapy. Unfortunately, the effect of TCT on the outcomes of HCC has not been explored. In this study, 7 public datasets and one external clinical cohort, including 1716 HCC patients were explored. Through WGCNA analysis and differential analysis, we explored the key TCT-related modulates. A total of 95 machine learning integrations across all validation cohorts were compared and the optimal method with the highest average C-index value was selected to construct the TCT derived signature (TCTS). In all independent clinical cohorts, TCTS showed accurate prediction of the prognosis, and was significantly correlated with clinical indicators and molecular features. Compared with 77 published gene signatures, the TCTS exhibited superior predictive performance. In the external clinical cohort, a novel nomogram (comprising TNM stage, Hepatitis B, Vascular invasion, Perineural invasion, AFP and TCTS) was constructed to test the clinical performance of TCTS. The results showed that the high TCTS scoring group showed dismal prognosis, improved sensitivity to oxaliplatin and good response to anti-PD-1/PD-L1 immunotherapy. Moreover, the low TCTS score group had few genomic alterations, low immune activation and low PD-1/PD-L1 expression levels. In conclusion, TCTS is an ideal biomarker for predicting the clinical outcomes and improving precision treatment of HCC.