Zhizhan Ni, Pengyuan Zhu, Li Liu, Haiyang Wang, Jinzhe Zhou, Xianqing Wang, Song Wang, Liangchen Zhu, Songze Zhang, Yongxin Zhou, Bujun Ge, Qi Huang, Chenshen Huang
ABSTRACT Gastrointestinal (GI) tumours account for a significant proportion of cancer cases and deaths globally; GI tumours show similarities regarding their development, diagnosis and treatment. Genomic and transcriptome analyses have been used to describe cancers, revealing driver mutations, aberrant regulatory pathways and several disease subtypes linked to tumours. However, these studies have not considered patient heterogeneity (intra‐tumour heterogeneity). Genetic heterogeneity, developmental programmes, epigenetics and the tumour microenvironment influence intra‐tumour heterogeneity. These factors are critical for tumour growth, clinical diagnosis, treatment resistance, metastasis and other cancer characteristics. Single‐cell genomics can evaluate transcriptional features at a single‐cell resolution and analyse genetic and functional heterogeneity. In the present study, we employed cNMF to illustrate intratumoral heterogeneity and identify shared programmes in scRNA‐seq and utilised novel machine‐learning algorithms to construct a signature based on the TCGA database and the different independent tumour‐type datasets from the GEO database in RNA‐seq. In addition, we evaluated whether this signature could predict survival, development, neoadjuvant therapy and drug efficacy. We employed the DepMap programme to identify the deficiency of genes in the constructed signatures that can affect tumour development.