MBA Education Management Center, UCAS

Recently, NPJ Precision Oncology published an article on the new AI model and algorithms applied to the identification of critical genes in colorectal cancer by Prof. Zhengjun Zhang at the University of Chinese Academy of Sciences (UCAS) and his team. The study found that four genes and their interactions identified from tissue samples can fully identify colorectal cancer. The four genes were cross-validated using a more rigorous cohort of over 2,000 cases from 10 cohorts of different populations with different study targets in different parts of the world. Previously, the role of these four genes has been sporadically reported in the literature, but the principle of their combined action has never been reported.

Prof. Zhang introduced that this new AI is a new model and algorithm that combines clustering and discrimination at the same time, and possesses the basic functions of the three major elements of deductive reasoning, inductive reasoning, and retrospective reasoning required for a real AI, and at the same time establishes the equivalence of biological significance and identified genes, i.e., it establishes invariance.

Currently, most AIs are "black boxes" with low interpretability, and can only satisfy one of the above three elements of inductive reasoning. The biggest feature of the new AI used in the study is its interpretability.

It is understood that the new AI algorithm can theoretically guarantee the existence and find the minimum core related genes. Since the objective function is a non-convex non-smooth function that combines combinatorial optimization and integer programming, the article proposes a new high-dimensional dimensionality reduction method based on the three major indices (mean, standard deviation, and Sharpe ratio) of the econometric model: a method of seeking differences and saving the same variable.

Further, the article proposes seven guidelines for identifying key variables that are more stringent than Hill's Criteria, which is commonly used in randomized experiments in clinical medicine.

Of the four genes identified, CXCL8/IL8 and PSMC2 were the ones for which relatively lower expression values were preferred, and SLC20A1 was the one for which relatively higher expression values were preferred, Prof. Zhang said. These three genes represent commonalities and consistency in colorectal cancer.

"The expression values of APP are heterogeneous. We found that the expression of this gene is retrograde in Europeans, Americans, Chinese, and Japanese. These features provide a new dimension of understanding and guidance for colorectal cancer diagnosis, test reagent development, drug development and treatment options." Prof. Zhang said that certain oncology hospitals in China provide patients with full genetic test reports that contain 500-600 genes, but the four genes they discovered are not in it, suggesting that the medical community does not know enough about colorectal cancer at the genomics level.

In view of this, it is pointed out that the heterogeneous expression of APP in Europeans and Americans and nationals especially needs to be paid attention to by the domestic medical community, especially the use of medication.

Zhengjun Zhang said that at present, about colorectal cancer associated genes achieving consistency has not been reported in published literature. Therefore, the new AI will certainly have a wide range of application scenarios. They expect to embed it into the underlying model of AI, and propose maximum linear regression and maximum logistic regression as the basic model of novel AI.

In addition to the study of colorectal cancer, genomic studies on the other four of the top five cancers (lung, breast, colorectal, liver, and stomach) released by the World Health Organization in terms of the number of people suffering have also been conducted.

To read the article ,please refer to https://doi.org/10.1038/s41698-024-00512-1