Table 1.

Various strategies that have been shown to predict immunotherapy outcomes with AI.

Prediction method	Forecast indicators	Outcomes	References
Liquid biopsy	Circulating tumor cell DNA, cytokines, serum complement levels, etc.	Circulating tumor cells can be used as a real-time detection system for targets such as PD-1; falling circulating tumor DNA was positively correlated with improved overall survival; serum C1q and LDH levels were correlated with the efficacy of immunotherapy	(45–47)
Multi-omics data	Genomics, transcriptomics, epigenomics, proteomics, radiomics, etc.	Multi-omics-based AI models present a chance to comprehend the information flow behind the disease. It can assess if each component of the model promotes the disease individually or whether they work together to treat it.	(48–50)
Clinical data	Population baseline data, medical history, examination results, etc.	Predictive models that distinguish immunotherapy age responders from non-responders can be constructed using data on patient age, sex, medical history, conventional laboratory tests, and follow-up CT scans	(51)
Tumor organoids	Tumor microenvironment, immune-tumor interaction, etc.	Organoids are very similar to the original tumor tissue, which can better mimic the in vivo immunotherapy response and observe the efficacy	(52, 53)
Others	miRNA abnormalities, gene mutations or recombination, gut microbes, etc.	The miRNA expression levels, ALK rearrangement, EGFR mutation, and gut microbial diversity were all related to the effect of anti-PD-1 treatment	(54–58)

AI, artificial intelligence; PD-1, programmed cell death protein 1; DNA, deoxyribonucleic acid; C1q, complement component 1, q subcomponent; LDH, lactate dehydrogenase; CT, computed tomography; miRNA, micro ribonucleic acid; ALK, anaplastic lymphoma kinase; EGFR, epidermal growth factor receptor.