Beyond KI67: Novel Biomarker Prediction for Tumor Growth Rates Using Machine Learning

The predominant characteristic of cancer is uncontrolled growth, and it has been demonstrated that tumor growth rate correlates with survival across a variety of tumor types. Measuring this growth rate directly necessitates evaluating the tumor size at two different points in time. Unfortunately, this is often impossible since tumors are typically removed shortly after diagnosis. An alternative is to assess the level of a biomarker that corresponds with the growth rate. A commonly used biomarker is the KI67 protein level. However, while KI67 levels serve as a suitable biomarker in healthy cells, they pose challenges in tumor samples due to frequent dysregulation in many types of tumors.
Here I would like to suggest a new approach to identifying an effective growth rate biomarker for tumors involves its derivation from the analysis of cancer cell lines, in which the growth rate can be easily measured. In this study, I thoroughly analyzed various types of omics data from 541 cancer cell lines sourced from the Cancer Cell Line Encyclopedia (CCLE), where their doubling times are known. Specifically, I sought mutated genes, copy number events, and gene RNA expressions that correlated with the doubling time. Notably, KI67 demonstrated only a weak correlation with doubling time, and more than 2000 genes exhibited a stronger correlation.